There is a shape to this that repeats. A business buys Copilot for 80 people. Week one, the usage numbers look healthy. By week six it has settled into about a dozen regulars, and most of that dozen were already the sort of people who try new software on a Saturday.
The read in the room is usually that staff are resistant, or too busy, or need more training. That has not been what I have found. The rollout finished at the licence and never reached the work.
In short
- Usage falls away because no one decided what Copilot was for in each role, not because people are unwilling.
- A single product demo teaches the tool. It does not teach the job.
- The questions that matter turn up two or three weeks after the session, by which point the trainer has gone.
- Adoption holds when someone owns it past go-live and the work gets picked before the features do.
The pattern
Licences get approved. IT enables them. An email goes out announcing that Copilot is now available, usually with a link to a Microsoft support page. There might be a lunchtime session.
Then everyone goes back to the work that was already on their desk.
What follows is predictable. People who were curious anyway keep using it. Everyone else opens it once, asks it something vague, gets something vague back, and quietly decides it is not for them. That decision is very hard to reverse. Six months later the renewal conversation starts with a finance question about cost per active user, and the honest answer is uncomfortable.
Where it actually breaks
No one decided what it was for
Copilot arrives as a capability rather than a job. That sounds like an advantage. In practice it means every person has to work out their own use case from scratch, in the middle of a week that is already full.
When the answer to "what should I use this for" is "anything", most people use it for nothing. The teams that get traction have three or four named tasks per role written down somewhere before the licences go live. Not twenty. Three or four, chosen because they happen every week and currently take too long.
The demo was impressive and irrelevant
Generic Copilot training tends to show the product at its best: summarise a meeting you were not in, draft an email to a person who does not exist, turn a document you have never read into slides.
Everyone nods. It looks clever. And none of it maps onto the report someone has to produce every Thursday from four spreadsheets that are formatted inconsistently, which is the task that would actually save them an afternoon.
Training that uses the client's own documents changes the room. People stop watching and start typing, because the example on screen is the thing sitting in their inbox.
Training happened once, and the questions came later
Week one questions are about buttons. Week three questions are the useful ones: why did it get this wrong, how do I stop it inventing a figure, can it see the SharePoint site, why does it work for my colleague and not for me.
By week three, most training engagements have finished. So the questions go unanswered, people draw their own conclusions about reliability, and usage drops.
No one owned it after go-live
Rollouts get a project owner. Adoption rarely does. Once the training is delivered and the invoice is paid, Copilot belongs to everyone, which means it belongs to no one in particular.
There is no one collecting the prompts that worked, no one deciding which team gets attention next, no one looking at the usage data and asking why finance is flat while marketing is not.
What closes the gap
Four things, roughly in order of how much difference they make.
Pick the work before the features. Sit with each team and find the recurring tasks worth attacking. The output is a short list per role, written in their language, not Microsoft's. This takes a couple of hours per team and it is the part people are most tempted to skip.
Train people in their own documents. Same session structure, different material. Bring the team's actual reports, actual client emails, actual decks. The learning transfers because there is nothing to transfer, they are already doing the job.
Give the questions somewhere to go. A standing 45 minutes a fortnight where people bring what broke is worth more than a second full training day. It catches the week-three questions while they still matter and it surfaces the patterns worth fixing for everyone.
Measure something a manager cares about. Licence utilisation tells you who opened the app. It does not tell you whether the Thursday report got faster. Pick two or three tasks at the start, note roughly how long they take, and check again at 30 days.
The uncomfortable part
Most of this is not about Copilot. It is about deciding what you want a team to do differently and then giving them enough support to actually do it, which is the same work any change takes.
The software has genuinely got better. The rollout method mostly has not.