All insights

Adoption · 16 May 2026 · 4 min read

Why Copilot training doesn't stick (and what does)

If your Copilot rollout has stalled three months in, you're not alone. Here's what we've learned about making training actually change how people work.

Author James Wilkinson

There is a pattern I see in nearly every firm I walk into.

Someone, usually six months ago, made the decision to buy Copilot. There was a kickoff. There were licences. There was probably a webinar (an hour, maybe ninety minutes) where the IT team or a vendor walked everyone through what Copilot can do. People left impressed. A few even went back to their desks and tried it out.

Then nothing.

By the time I get the call, usage data looks the same as it did the week the licences turned on. The people who were going to be early adopters already are. Everyone else has quietly gone back to working the way they always did. The leadership team is starting to ask, gently at first, whether they made a mistake.

They didn’t make a mistake buying it. They made a mistake in what they did next.

The “train and leave” trap

The phrase I use for this is train and leave. You deliver a session. You hand over a quick-reference PDF. You say “if you have any questions, raise a ticket.” Then you call it done.

It fails for a specific reason, and it is not the reason most people think. It is not that the training was too short, or that the slides were ugly, or that people don’t care. It is that AI does not behave like the other software your team uses.

When you train someone on a new expenses system, the system has a finite number of buttons. They learn the buttons. They use the buttons. The training maps cleanly to the work.

Copilot does not have buttons in any meaningful sense. The interface is a blinking cursor and the promise that, if you ask the right question in the right way with the right context attached, something useful will happen. That is a fundamentally different skill from clicking a button. And it cannot be taught in ninety minutes any more than a language can.

What actually changes behaviour

Three things, in my experience, move the needle. None of them are particularly clever, which is part of why they are so often skipped.

One: training has to be tied to real work that real people actually do this week. The fastest way to kill engagement is a demo where Copilot summarises a fake email about a fake project. Nobody cares about the fake email. They care about the email sitting in their inbox right now from the client who is unhappy. Train on that. Sit with people. Use their files, their tone, their actual deliverables. The first time someone watches Copilot draft something that sounds like them, the lights come on.

Two: there has to be a gap between sessions and the gap has to have homework. Not a homework sheet. Homework as in “between now and our next session, find one thing in your week you would normally spend 30 minutes on and try it with Copilot first.” Then they bring back what worked, what didn’t, and what they got stuck on. That conversation is where the actual learning happens. It is also where you find the use cases nobody would have thought of in a strategy meeting.

Three: somebody in the room has to keep doing this after you leave. A champion, an ambassador, a power user. The label matters less than the function. They are the person someone can grab in the kitchen and ask “is there a faster way to do this?” Without that role embedded somewhere, every question becomes an IT ticket, and IT tickets are where curiosity goes to die.

The honest version

There is no version of this that is fast and cheap and easy. AI training that sticks looks like a programme, not an event. It runs over weeks, not hours. It involves the messy work of sitting with people and watching them try things and getting it wrong and trying again.

The firms that get the most out of Copilot are the ones who treat the rollout as a culture change project that happens to involve software, rather than a software project that happens to involve people. That is the order, and it matters.

If your rollout has stalled, the answer is not another all-hands webinar. The answer is to go back to the smallest group of people who actually want this to work, find out what they are trying to do, and start there.

That is the shape of Copilot training that sticks: practical cohorts, live online delivery through Microsoft Teams where useful, webinars for wider awareness and enough follow-up for new habits to survive launch week.