04_BUSINESS_AUTOMATION_SYSTEMS

LAK CAPITAL | PREMIUM PRACTICAL FORMATION COMPLETE PROFESSIONAL FORMATION MODULE 04: BUSINESS AUTOMATION SYSTEMS Automate support, email, content and core business operations. REAL SYSTEMS • PRACTICAL DELIVERY • PROFESSIONAL RESULTS.

AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 2 MODULE 04 | REAL PRACTICAL DEMONSTRATION SUPPORT TRIAGE SYSTEM Automate support, email, content and core business operations. REAL INPUT EXACT BUILD ACTION VERIFIED OUTPUT Input: customer ticket, account tier, product area and knowledge articles. Classify intent, retrieve approved knowledge, draft a response and escalate billing, legal or low-confidence cases. Output: faster first response without allowing the model to invent policy. SYSTEM VIEW TRIGGER VALIDATE DECIDE ACT PROVE Professional rule: Never activate a production action until normal, incomplete, ambiguous, duplicate and failure cases have been tested and recorded..

AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how LAK CAPITAL | AI AUTOMATION AGENCY 3.

AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 3 Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate If the result is wrong first inspect the LAK CAPITAL | AI AUTOMATION AGENCY 4.

AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 4 p p 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 p p p 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 p p p 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 p p p 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 4 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Create three courses for high, medium and low scores. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 13 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to speed up the follow-up without degrade the customer relationship., to cite the necessary data, to describe the case that requires human validation and to show in the history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 5.

AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 5 ORIGINAL PROFESSIONAL UNIT 14 Module 14 Automate Customer Support Answer quickly while knowing how to climb. Jurisdictions • Sort requests and emergencies • Seek a procedure and generate a well-founded response • Use trust, ALS and human transfer Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 5 ORIGINAL PROFESSIONAL UNIT 14 Module 14 Automate Customer Support Answer quickly while knowing how to climb. Jurisdictions • Sort requests and emergencies • Seek a procedure and generate a well-founded response • Use trust, ALS and human transfer Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 5 ORIGINAL PROFESSIONAL UNIT 14 Module 14 Automate Customer Support Answer quickly while knowing how to climb. Jurisdictions • Sort requests and emergencies • Seek a procedure and generate a well-founded response • Use trust, ALS and human transfer Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 14 Module 14 Automate Customer Support Answer quickly while knowing how to climb. Jurisdictions • Sort requests and emergencies • Seek a procedure and generate a well-founded response • Use trust, ALS and human transfer Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 14 Module 14 Automate Customer Support Answer quickly while knowing how to climb. Jurisdictions • Sort requests and emergencies • Seek a procedure and generate a well-founded response • Use trust, ALS and human transfer Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 14 Module 14 Automate Customer Support Answer quickly while knowing how to climb. Jurisdictions • Sort requests and emergencies • Seek a procedure and generate a well-founded response • Use trust, ALS and human transfer Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 14 Module 14 Automate Customer Support Answer quickly while knowing how to climb. Jurisdictions • Sort requests and emergencies • Seek a procedure and generate a well-founded response • Use trust, ALS and human transfer Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 14 Module 14 Automate Customer Support Answer quickly while knowing how to climb. Jurisdictions • Sort requests and emergencies • Seek a procedure and generate a well-founded response • Use trust, ALS and human transfer Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 14 Module 14 Automate Customer Support Answer quickly while knowing how to climb. Jurisdictions • Sort requests and emergencies • Seek a procedure and generate a well-founded response • Use trust, ALS and human transfer Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 14 Module 14 Automate Customer Support Answer quickly while knowing how to climb. Jurisdictions • Sort requests and emergencies • Seek a procedure and generate a well-founded response • Use trust, ALS and human transfer Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 14 Module 14 Automate Customer Support Answer quickly while knowing how to climb. Jurisdictions • Sort requests and emergencies • Seek a procedure and generate a well-founded response • Use trust, ALS and human transfer Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: LAK CAPITAL | AI AUTOMATION AGENCY 6.

AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 5 Lesson 1 Understanding the Principle In this module, you will learn how to classify requests and emergencies. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, such as a free email, document, question or customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to search for a procedure and generate a based answer. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to use trust, sla and human transfer. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: respond quickly while knowing how to climb.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model only deals with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate If the result is wrong first inspect the LAK CAPITAL | AI AUTOMATION AGENCY 7.

AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 6 p p 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 p p p 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 p p p 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 p p p 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 6 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Build an assistant support covering five ticket categories. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 14 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to respond quickly while knowing how to climb., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, redo the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8.

AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 7 ORIGINAL PROFESSIONAL UNIT 15 Module 15 Automate Email Management Transform an inbox into a structured work queue. Jurisdictions • Sort, summarize and extract data • Generate drafts under control • Create Tasks, CRM Updates and Alerts Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 7 ORIGINAL PROFESSIONAL UNIT 15 Module 15 Automate Email Management Transform an inbox into a structured work queue. Jurisdictions • Sort, summarize and extract data • Generate drafts under control • Create Tasks, CRM Updates and Alerts Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 7 ORIGINAL PROFESSIONAL UNIT 15 Module 15 Automate Email Management Transform an inbox into a structured work queue. Jurisdictions • Sort, summarize and extract data • Generate drafts under control • Create Tasks, CRM Updates and Alerts Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 15 Module 15 Automate Email Management Transform an inbox into a structured work queue. Jurisdictions • Sort, summarize and extract data • Generate drafts under control • Create Tasks, CRM Updates and Alerts Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 15 Module 15 Automate Email Management Transform an inbox into a structured work queue. Jurisdictions • Sort, summarize and extract data • Generate drafts under control • Create Tasks, CRM Updates and Alerts Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 15 Module 15 Automate Email Management Transform an inbox into a structured work queue. Jurisdictions • Sort, summarize and extract data • Generate drafts under control • Create Tasks, CRM Updates and Alerts Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 15 Module 15 Automate Email Management Transform an inbox into a structured work queue. Jurisdictions • Sort, summarize and extract data • Generate drafts under control • Create Tasks, CRM Updates and Alerts Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 15 Module 15 Automate Email Management Transform an inbox into a structured work queue. Jurisdictions • Sort, summarize and extract data • Generate drafts under control • Create Tasks, CRM Updates and Alerts Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 15 Module 15 Automate Email Management Transform an inbox into a structured work queue. Jurisdictions • Sort, summarize and extract data • Generate drafts under control • Create Tasks, CRM Updates and Alerts Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 15 Module 15 Automate Email Management Transform an inbox into a structured work queue. Jurisdictions • Sort, summarize and extract data • Generate drafts under control • Create Tasks, CRM Updates and Alerts Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 15 Module 15 Automate Email Management Transform an inbox into a structured work queue. Jurisdictions • Sort, summarize and extract data • Generate drafts under control • Create Tasks, CRM Updates and Alerts Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, LAK CAPITAL | AI AUTOMATION AGENCY 9.

AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 7 Lesson 1 Understanding the Principle In this module, you will learn how to classify, summarize and extract data. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to generate drafts under control. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how it has been transformed and what data has been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to create tasks, crm updates and alerts. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to transform a Inbox into a structured work file.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate If the result is wrong first inspect the LAK CAPITAL | AI AUTOMATION AGENCY 10.

AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 8 p p 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 p p p 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 p p p 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 p p p 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 8 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Treat twenty shadow emails with privacy rules. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 15 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the goal is to turn a Inbox into a structured work queue., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 11.

AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 9 ORIGINAL PROFESSIONAL UNIT 16 Module 16 Automate Content Production Industrize production without sacrificing quality. Jurisdictions • Separate research, strategy, generation, review and publication • Apply voice, editorial rules and SEO • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 9 ORIGINAL PROFESSIONAL UNIT 16 Module 16 Automate Content Production Industrize production without sacrificing quality. Jurisdictions • Separate research, strategy, generation, review and publication • Apply voice, editorial rules and SEO • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 9 ORIGINAL PROFESSIONAL UNIT 16 Module 16 Automate Content Production Industrize production without sacrificing quality. Jurisdictions • Separate research, strategy, generation, review and publication • Apply voice, editorial rules and SEO • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 16 Module 16 Automate Content Production Industrize production without sacrificing quality. Jurisdictions • Separate research, strategy, generation, review and publication • Apply voice, editorial rules and SEO • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 16 Module 16 Automate Content Production Industrize production without sacrificing quality. Jurisdictions • Separate research, strategy, generation, review and publication • Apply voice, editorial rules and SEO • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 16 Module 16 Automate Content Production Industrize production without sacrificing quality. Jurisdictions • Separate research, strategy, generation, review and publication • Apply voice, editorial rules and SEO • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 16 Module 16 Automate Content Production Industrize production without sacrificing quality. Jurisdictions • Separate research, strategy, generation, review and publication • Apply voice, editorial rules and SEO • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 16 Module 16 Automate Content Production Industrize production without sacrificing quality. Jurisdictions • Separate research, strategy, generation, review and publication • Apply voice, editorial rules and SEO • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 16 Module 16 Automate Content Production Industrize production without sacrificing quality. Jurisdictions • Separate research, strategy, generation, review and publication • Apply voice, editorial rules and SEO • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 16 Module 16 Automate Content Production Industrize production without sacrificing quality. Jurisdictions • Separate research, strategy, generation, review and publication • Apply voice, editorial rules and SEO • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL ORIGINAL PROFESSIONAL UNIT 16 Module 16 Automate Content Production Industrize production without sacrificing quality. Jurisdictions • Separate research, strategy, generation, review and publication • Apply voice, editorial rules and SEO • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your LAK CAPITAL | AI AUTOMATION AGENCY 12.

AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 9 In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 g p In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 g p In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice LAK CAPITAL | AI AUTOMATION AGENCY 9 • Maintaining Human Validation Lesson 1 Understanding the Principle In this module, you will learn how to separate research, strategy, generation, review and publication. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to apply voice, editorial rules and soo. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to maintain a human validation. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: industrialize production without sacrificing quality.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Open your automation tool and create a new empty workflow. 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Open your automation tool and create a new empty workflow. 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Open your automation tool and create a new empty workflow. 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action if the incomplete case is stopped AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Open your automation tool and create a new empty workflow. 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action if the incomplete case is stopped AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Open your automation tool and create a new empty workflow. 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action if the incomplete case is stopped AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Open your automation tool and create a new empty workflow. 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action if the incomplete case is stopped AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Open your automation tool and create a new empty workflow. 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action if the incomplete case is stopped AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Open your automation tool and create a new empty workflow. 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action if the incomplete case is stopped AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Open your automation tool and create a new empty workflow. 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action if the incomplete case is stopped AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Open your automation tool and create a new empty workflow. 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Y i i t if th l h th fi l ti if th i l t i t d AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Open your automation tool and create a new empty workflow. 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Y i i t if th l h th fi l ti if th i l t i t d AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL 1. Open your automation tool and create a new empty workflow. 1. Give the workflow a name that describes the job result, not the name of a tool. 1. Add a manual trigger to be able to test safely. 1. Create three fictitious data: normal, incomplete and ambiguous. 1. Add a transformation step and clearly rename its fields. 1. Add a simple condition that separates the accepted case from the case to be checked. 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed i i if h l h h fi l i if h i l i d LAK CAPITAL | AI AUTOMATION AGENCY 13.

AI AUTOMATION AGENCY | COMPLETE FORMATION LAK CAPITAL LAK CAPITAL | AI AUTOMATION AGENCY 3 ORIGINAL PROFESSIONAL UNIT 13 Module 13 Automate Business Processes Accelerate follow-up without degrade customer relationship. Jurisdictions • Qualifying Fit, Need, Budget and Intent • Customize Recoverys • Conducting by score and measuring conversion Lesson 1 Understanding the Principle In this module, you will learn to qualify fit, need, budget and intent. Always start with a concrete situation: a person receives information, the interpreter, makes a decision and then performs an action. Automation reproduces this chain. AI becomes useful when the information is not perfectly structured, for example a free email, a document, a question or a customer comment. A standard rule is appropriate for the exact decisions: if the amount exceeds 10,000, ask for approval. An AI template is more suitable for an understanding task: to determine whether a message expresses a claim, to summarize a document or to extract a name. Do not entrust the model with a rule that can be clearly written. This separation makes the system more reliable, less costly and easier to explain. Lesson 2 Observing data The second skill is to customize the re-launches. Click on the history of an execution in your tool and open the output of each step. You must be able to answer three questions: what data is entered, how they have been transformed and what data have been released. If you do not see the expected value, the problem is usually in the mapping, format or previous step. Make it a habit to clearly name the fields: email client is better than value1. An ID must remain stable. A date must be used in an explicit format. A status must belong to a short and known list. This discipline avoids most errors when workflow becomes longer. Lesson 3 Building without getting lost The third skill is to route according to the score and measure conversion. Build one step at a time. Run it, control its output and add the next. First use simple fictitious data. Then add an incomplete example, an ambiguous example and an incorrect example. A system is not ready because it works once; it is ready when its behavior remains predictable in several situations. Example explained Suppose the objective is: to speed up the follow-up without degrade the customer relationship.. The trigger receives an input. A preparation step cleans the fields. Fixed logic eliminates impossible cases. The AI model deals only with the part that requires understanding language or context. A rule then checks the format of the response. Finally, the system executes the authorized action and writes a journal. If confidence is weak, it creates a task for a person instead of acting alone. Step by step guided practice 1. Open your automation tool and create a new empty workflow. LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 p p j 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 p p j J 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 , p p j , , J 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 10 1. If the module uses AI, write a short prompt with objective, data, rules and JSON format. 1. Run each step separately and read all its output. 1. Add the final action in test or draft mode. 1. Add an error path that records the problem and alerts a person. 1. Re-launch all three cases and compare the expected result to the result achieved. 1. Correct one cause at a time, then repeat the test. Module application Produce a week of content from a single brief. Correction and result to be observed Your scenario is correct if the normal case reaches the final action, if the incomplete case is stopped with an understandable message and if the ambiguous case requires validation. Also check that an identical second execution does not silently create a duplicate. If the result is wrong, first inspect the data transmitted before changing the AI statement. Frequent errors • Build the entire scenario before executing the first step. • Send all data to the model while some are useless or sensitive. • Ask for a free answer and then try to use it as a structured data. • Test only a perfect example. • Enable a real action before testing the error path. Check that you understand • Can you explain in two minutes the objective of module 16 without mentioning a tool? • What data are essential and which are optional? • What failure is most likely and how will it be detected? • What stage requires human validation? Due replies You must be able to explain that the objective is to industrialize production without sacrificing quality., to cite the necessary data, to describe the case that requires human validation and to show in history how the workflow has made its decision. If you cannot explain it simply, re-explain the practice with only three steps. LEVEL 3 METAL AUTOMATOMATIZATION LAK CAPITAL | AI AUTOMATION AGENCY 14.