Microsoft 365 Copilot Adoption: Why Seats Sit Unused

Microsoft had sold over 20 million Microsoft 365 Copilot paid seats by its third quarter of financial year 2026, with the number of customers buying more than 50,000 seats quadrupling year on year. Buying the licence is the easy part. Getting people to use it, well, every day, is a separate problem, and it is not solved by the licence itself.

The size of the problem

Gartner’s own June 2026 assessment of Microsoft 365 Copilot put it plainly: “wide-scale adoption and value realisation is not guaranteed.” It matches what most IT and L&D teams already sense from their own usage reports: a purchase order and a login are not the same thing as a habit.

Why buying a licence does not create a user

A licence gives someone access to Copilot. It does not tell them what to type into it, when it is worth opening instead of doing the task by hand, or how to check what it gives back before sending it on. Without that, most people try it once or twice on something unfamiliar, get an answer that needs more editing than starting from scratch, and go back to their old way of working. The tool gets blamed for a training gap.

The pattern repeats across most rollouts we hear about from buyers: IT enables the licence for everyone in a department in one go, an email goes round with a link to Microsoft’s own help pages, and usage spikes for a week before dropping back close to where it started.

What Microsoft’s own research says

Microsoft’s 2026 Work Trend Index found that 67% of the variance in whether AI investment pays off comes from organisational factors, chiefly manager behaviour, culture and how work is structured, against 32% linked to individual mindset or capability. Its own conclusion was direct: managers who visibly use AI themselves, and who encourage experimentation, are among the strongest predictors of whether their teams get real value from it.

That points at exactly where training earns its keep: a course built for managers, not just for individual contributors, changes the behaviour that Microsoft’s own data says matters most. If a manager has sat through a course that shows Copilot doing something recognisably part of their own job, they are far more likely to use it visibly, and that visible use is what spreads.

Why generic AI training does not fix a Copilot problem

A general course on what AI is and how to prompt it well is useful, but it will not tell a finance manager how to draft a variance commentary in Excel with Copilot, or a sales lead how to prepare account research in Outlook and Teams before a call. Copilot lives inside specific, everyday tools, and the training that changes behaviour has to live there too.

What training should not look like

A single, generic hour that shows the same three prompts to every department is the version most likely to fail, for the same reason a single Article 4 training session fails: it cannot show a finance manager and a salesperson the parts of Copilot that touch their own work. A recorded webinar with no pass mark has the same weakness from a different angle: it proves someone opened a link, not that they came away able to do anything differently on a Tuesday afternoon.

What actually closes the gap

Three things, in order:

• Role specific training, not one course for the whole organisation, such as our Microsoft 365 Copilot training. A manager, a salesperson and someone in finance or HR use Copilot for different tasks and need to see their own tasks in the course.
• Assessed, not just watched. A course with a pass mark produces a completion you can point to; a video nobody has to finish does not tell you whether anyone learned anything.
• Training for managers specifically, such as Microsoft 365 Copilot for Managers & Leaders, because Microsoft’s own data says their visible use of the tool is the biggest single factor in whether their team adopts it.

How to measure whether it worked

Usage reports inside Microsoft 365 already show who has opened Copilot in the last 28 days, by department. Pull that report before training runs, run the role specific courses, then pull the same report again a month later. A rise in weekly active use within the trained group, set against a flat line in a group that has not yet been trained, is a clean before and after that does not depend on anyone’s opinion of whether the training felt useful.

It is also worth tracking who has finished the training, not just who has opened Copilot. An assessed course gives you that record directly: a completion and a pass mark against each learner’s name, which you can set alongside the usage report to see whether the two actually move together in your organisation.

What this looks like in practice

A short, practical course for everyone who has been given a seat, about 45 minutes, covering what Copilot is and the habits that make it worth using. Then a separate, equally short course for managers and leaders, because their behaviour carries more weight than anyone else’s. Then, where it is worth the investment, a course for a specific function such as sales, finance or HR, each showing Copilot doing that team’s actual work rather than a generic example. Twenty million seats is a large number. What decides whether it was money well spent is not the number of seats, it is how many of them get opened on a Tuesday afternoon for a real task.