The Training Was a Success. Three Weeks Later, Nothing Had Changed.

A team spends a day learning to use AI well. People leave energised, the feedback is warm, the manager is pleased. Three weeks later, you ask who has used what they learned. A few hands go up. The rest say they meant to, but the week got busy.

Nobody failed here. The training worked. The adoption did not happen. Those are two different things, and most companies only budget for the first.

What is actually happening

Training transfers knowledge. Adoption requires a change in how a Tuesday at 3pm works. Back at their desks, people meet the same inbox, the same deadlines, the same templates and the same manager expectations as before. Nothing in that environment asks them to behave differently.

Using AI starts with slowing down: choosing a task, writing a decent prompt, checking the output, fixing it. Under pressure, the old way is faster, so the old way wins.

There is also a quieter problem. The examples used in training were good but generic. Each participant has to translate them to their own job alone, with no one to ask. Translation is where most people stop.

The reframe

Many leaders conclude that they need more training, better tools or more motivated people. What actually matters is whether the work itself has been redesigned to include AI. Adoption is not a skills gap, it is a workflow gap. Skills are necessary, but without a place in the workflow they fade within weeks.

In MAKIA terms, this is where Meaning and Actors matter: does each person know which of their tasks AI is for, and does someone own that conversation after the workshop ends?

What this means for an SME

1. Choose the task before the tool. For each team, name two or three recurring tasks where AI becomes the default starting point: the weekly report, the first draft of a client reply, the meeting summary. If nobody can name them, that is already the real finding.
2. Give the follow through an owner. Not a champion with a title, just one person per team who asks the same question every week: what did we try? Ten minutes in an existing meeting is enough.
3. Protect the slower first attempt. Say explicitly that the first AI assisted version of a task may take longer. If deadlines never bend, adoption never starts.
4. Share the ugly examples. Prompts that failed and outputs that needed rewriting teach more than polished demos, and they make it safe to try.

To close

If your last AI training ended with no measurable change, ask a different question than "was it good?". Ask: which task in our week now looks different? If the answer is none, the training is not the thing to fix.

Try this week

Take 25 minutes. Open your team's calendar for the past two weeks and spot one task that comes back every week. On a single page, write down: the task, who does it, how long it takes today, and the step where AI could produce the first draft. Then ask one person to try it once this week, and to tell you what happened in ten minutes at your next team meeting.