Training your teams in AI: what's left three months later
Four observations on AI training in companies. The first explains the urgency, the other three explain why existing programmes don't address it.
In this article5 sections
In most companies, AI is already being used, with no framework
Recent surveys put the share of employees using AI tools not approved by their employer at between half and two thirds, and executives are among the heaviest users.
These tools receive internal documents, client files, sometimes HR data.
The risk isn't that your teams start using AI: they already are. The risk is that nobody can say what has left the building.
Adoption is often declared and rarely measured
Hundreds of corporate deployments have been examined: the overwhelming majority of pilots produced no measurable effect on results.
The cause isn't the quality of the models: it's that the tool was never connected to real work.
A subscription paid for the whole company, a few enthusiasts, and nothing changing in the processes: it's the most common situation, and it's expensive.
A training day produces awareness, not a habit
Everyone comes out motivated, and three weeks later the workload has taken over again.
Research on learning transfer is consistent: without structured follow-up, only a small minority of what was learned ends up being applied.
It's not a problem of motivation or the trainer's quality: changing a way of working takes repetition, and a one-off event offers none. A second short session a few weeks later, on the cases that have got stuck since, is often worth more than lengthening the first.
General AI training changes nothing about a particular job
It's the toughest finding of recent surveys: the vast majority of trained employees say what they learned doesn't help them in their job.
They were shown what a language model is and how to write a prompt. Nobody showed them how to process their invoices, write their proposals, prepare their minutes.
The gap between the demo and Tuesday morning was never crossed. It's crossed by working on three or four real tasks, with real documents, rather than textbook cases.
The usage framework is written with the people who will apply it
Which data can go into an external tool, which never leaves, which tools are approved, who reviews what before sending.
One page, decided with your teams rather than imposed by a memo. A rule written by those who apply it is the only one that holds, and it brings into the frame the uses that already existed on the quiet.
It's worth also naming the tasks where AI is useless or dangerous in your company. Knowing where not to use it is part of the training.
AI training isn't judged on the day, when everyone is happy. It's judged a few weeks later, on what has been kept and what has been dropped, and that's precisely the part most programmes don't do.
Ce sujet touche à notre expertise Workshop & Training.
We start with a conversation, not a quote.
Voir l'expertiseRead next
