Training your teams in generative AI: the standard three-day programme
Three days are enough to make a business team self-sufficient with generative AI, provided they work on their own tasks. Here is the outline we follow.
A useful generative AI course is judged by what the participants produce, not by what they listened to. Our three-day outline devotes the first day to understanding the real limits, the second to writing reliable instructions on the company's own documents, and the third to building the roadmap.
Why three days and not half a day?
Half a day produces enthusiasm, then nothing. Participants leave with the impression that the tool is magic, run into the first wrong answer, and give up.
Three days allow the full cycle: understand, practise on real material, get it wrong, correct, and leave with something that works.
What does each day contain?
| Day | Content | Deliverable |
|---|---|---|
| 1 | What the model really does, its typical failures, the confidentiality issues | List of candidate tasks for the department |
| 2 | Writing instructions, breaking down tasks, checking answers | Instructions tested on your own documents |
| 3 | Mapping by department, choosing between an online tool and an internal deployment | Three-month roadmap |
Are technical prerequisites needed?
None. The audience is precisely those who will never write code: management, administration, human resources, quality, sales.
One non-technical prerequisite is essential, however: participants must arrive with real tasks from their daily work. Without material, the course stays theoretical.
Which tools do we work on?
Those the company already uses or is considering. Method matters more than tool, and the principles carry across from one assistant to another.
Where the company handles sensitive data, the third day covers deploying an internal model, which avoids sending its documents outside.
How do you know whether the course worked?
Set the measure before you start. How many hours a week does the target task consume today? Ask the question again six weeks later.
That is the only criterion worth having. A satisfaction score taken on the day measures the mood in the room, not the effect on the organisation.
What are the most frequent mistakes?
Training everybody at once without identified use cases. Banning the tool without offering an alternative, which pushes teams to use it uncontrolled from their phones. And confusing an impressive demonstration with a real gain.
Updated Aug. 11, 2026