AI in customer service
M29
Ground answers (RAG), mask data, escalate to a human.
Online10–15 min
Training by function
Customer service
A virtual agent speaks on behalf of the company. This path teaches how to ground its answers, mask sensitive data and plan escalation to a human.
Why this path
A virtual agent's words bind the company.
An ungrounded assistant invents policies, promises refunds or discloses another client's data. The defense: ground answers in a verified knowledge base (RAG), mask sensitive information and guarantee a handoff to a human. This path installs those guardrails.
The module dedicated to your function
The common core for everyone
The baseline reflexes your whole team shares, whatever the job.
Understanding generative AI
M01
Demystify the probabilistic engine and distinguish it from predictive AI.
Online10–15 min
A probabilistic engine predicts the most plausible word without understanding meaning: brilliant, but sometimes wrong. Never feed it confidential data in a public tool.
RegisterHallucinations
M02
Identify fabricated information and how to verify it at the source.
Online10–15 min
A hallucination is fabricated information presented confidently. The risk is real on expert tasks (figures, citations): every output must be verified at the source.
RegisterUsing an assistant well
M03
Structure your prompts (Role + Context + Task + Format) and iterate.
Online10–15 min
Good results come from a structured prompt and an iterative approach. Providing examples of the desired output markedly improves quality.
RegisterConfidential data: the basic reflexes
M04
Sort your data and de-identify rigorously before every prompt.
Online10–15 min
Anything entered into a public AI can be retained and reused. Never client names, salaries or API keys; removing just the name is not enough.
RegisterApproved tools and "shadow AI"
M05
Recognize shadow AI and have tools approved by IT.
Online10–15 min
Shadow AI is the use of unapproved tools, often free and poorly protective. The reflex: check the IT allowlist and get the tool approved before using it.
RegisterYou remain responsible: the human role
M06
Apply "human-in-the-loop" and the "four eyes" method.
Online10–15 min
AI is a gifted junior assistant with no judgment: final responsibility stays human. Review and validate every deliverable as if you had written it yourself.
RegisterGo deeper for your context
The modules most relevant to the risks specific to your function.
Confidential data: the basic reflexes
M04
Sort your data and de-identify rigorously before every prompt.
Online10–15 min
Anything entered into a public AI can be retained and reused. Never client names, salaries or API keys; removing just the name is not enough.
RegisterTransparency: when and how to disclose AI use
M18
Know when disclosure is required and how to phrase it.
Online10–15 min
Inform people they are interacting with an AI and own its use while affirming human validation. For low-level assistance, disclosure remains optional.
RegisterAnonymization vs pseudonymization
M08
Distinguish reversible masking from legal, irreversible anonymization.
Online10–15 min
Masking a name is not enough: indirect clues (postal code, role, unique detail) allow re-identification. Legal anonymization is irreversible.
RegisterReady to equip your team?
Enroll your teams in our AI-literacy and governance training paths, from the common core to function-specific use cases.
Immediate access · 100% online · at your own pace