Training a team on AI in one day

An owner who wants to train a team often asks me the same question: in one day, what can people actually learn? The answer depends less on the tool than on what goes into the day. Here is how I build it, from what I did in Toulouse with PREVALY, an occupational health and prevention service, and in Réunion with the CPME.

Starting from each person’s level

Before the content, I ask who is in the room. How long each person has been using these tools: three months, six months, a year. Every day, or once a week. For what: rewriting an email, building a plan, something else. I introduce myself afterwards.

That round is not a courtesy. In the same team, the gaps are wide. Some already use it every day; others have never opened it, or closed it again after a bad answer. A day that ignores that gap speaks to part of the room and loses the rest.

A brief theoretical introduction

The theory opens the day, but it is not the heart of it. I set down the minimum needed to avoid picking the wrong tool.

A language model produces the most coherent continuation of what it was given. It is not built to flag, on its own, that it does not know. A good share of the errors comes from there, and that is why you read the result over.

I also show what you do not see on the screen: the instructions the vendor set upstream, the personalisation, the memory, and how much text the tool can read at one time. These are not technical details. They explain why two people do not get the same answer to the same question.

The conclusion comes quickly: you test it. And when you do not know how to phrase an instruction, you can ask the tool.

Then, the team’s documents

The rest of the day is done on the team’s own papers, not on an example of mine.

We create a project in the tool and load working documents into it, anonymised if needed. The image I use is a drawer: the papers stay in the project, and every conversation in the project can go back to them. You avoid copying everything in again each time.

I have it analyse first, then write. If you ask it to write straight away, the tool leans on the whole pile, with no hierarchy. If you ask for an analysis first, the writing that follows leans on that analysis. I also show how to structure a request, with headings and documents clearly separated: for the tool, everything sits at the same level until you split it.

The point is to see, on a file the team knows, what the tool already produces, and what has to be read over. That is where each person measures the value and the limit of the tool for their own work.

In Toulouse, the practice took the form of deliverables. The team wanted materials for the firms it supports. We started from what already existed, a self-assessment, a card game, an FAQ, with the aim of leaving with a few first versions. In the afternoon, we worked on an interview guide. Once the guide is ready, a questionnaire is derived from it: you give it to the tool and ask for a version that people can complete on their own. You start with the hardest piece; the others come faster.

What you can do when you leave

When you leave, you have not “learned AI”. You know, roughly, what a language model does, and why you have to read the result over. You have loaded documents into a project, asked for an analysis, then a draft, and read the result against work you know.

Without that rereading, the day produced a text, not a skill. And you leave with first versions, not with finished tools.

What to plan for afterwards

A day does not hold on its own. What was seen in the room fades if no one uses it the following week.

In Toulouse, the day opened three months of support: a check-in planned a month and a half later, another around three months, and an open channel for questions in between. What I take from it: waiting for questions is not enough. It is better to come back to the team at regular intervals, on its own documents, and to adjust to each person’s actual use.

That is also what participants say. Maryse, after an AI training session at CPME Réunion: “This training session went very well, was of good quality and very useful. A discovery for me, as I am not familiar with this tool. Well done!”

The format

The training is built on the team’s own cases, at €1,500 excl. VAT per day, with a half-day available. It is not Qualiopi-listed, by choice: it therefore cannot be funded through professional training funds (OPCO, CPF). The scope and the price are written down before work starts.

Training in Réunion is on this page. For an SME, in Réunion or in mainland France, it is this one.