Your Copilot business case rests on a survey. The one department that timed it found something else.

Your Microsoft renewal is due. The account manager has a proposal to put Copilot on every seat, and a figure to go with it: staff save around half an hour a day. Somebody has worked out what half an hour a day is worth across 400 people, and the number is large enough that the licence cost looks like a rounding error.

Ask where the half hour came from.

In almost every case it came from asking people how much time they think they saved. That is not nothing. It is also not a measurement, and the one serious attempt in the UK to check it against a stopwatch did not entirely agree.

Three government trials, one product, one quarter

Between September and December 2024, three parts of the UK government trialled Microsoft 365 Copilot at scale. All three have now published what they found, which makes them the best public evidence a UK buyer has.

The cross-government experiment gave licences to 20,000 staff across 12 organisations and reported an average saving of 26 minutes a day. The report says the figures were "self reported": participants picked a range on a survey, and the average was calculated from the midpoints. 7,115 people answered.

HMRC's evaluation, published in July 2026, randomly allocated 3,000 licences. Its staff reported saving 2 to 3% of their working week, about an hour a week, after the analysts cut the raw survey answers by around 20% to correct for keen users being more likely to respond. HMRC says the figure "should be treated as a baseline estimate rather than a precise measure of impact".

The Department for Business and Trade gave 1,000 staff licences and concluded: "We did not find robust evidence to suggest that time savings are leading to improved productivity."

Same product, same three months, and the headline benefit ranges from about 12 minutes a day to 26, with a third department unable to find it turning into anything at all. All of these numbers come from people's own estimates.

To be fair to DBT, it says productivity was not a main aim of its evaluation, so it collected little data on it. The honest summary is not "Copilot does nothing". It is that nobody in these trials measured what you are about to pay for.

What happened when somebody timed it

DBT did one thing the others did not. It sat people down with real tasks, timed them, and had a second analyst blind-mark the results, comparing Copilot users with colleagues who had no licence.

The sample was small, 11 users and 5 non-users, so treat it as a warning light rather than a verdict. What it showed:

• Summaries and emails were faster and better with Copilot, though the time saved on emails was, in the report's words, "extremely small".
• PowerPoint slides came out over seven minutes faster, and worse. Lower quality and lower accuracy than non-users managed.
• Excel data analysis was slower and worse. This directly contradicted the diary study, where users had reported saving time on data analysis.

That last one is the finding to sit with. On the task where a mistake is most expensive, people believed they were saving time, and when somebody watched, they were not.

The same report found that across the pilot the average user took 1.14 Copilot actions a day, and about 30% used it at least once each working day. 22% of diary respondents said they had spotted it making things up; another 22% declined to answer that question.

Is Microsoft Copilot worth it for a UK business?

For some roles, probably. All three trials found the clearest value in written work: summarising meetings and documents, drafting, tidying up prose. The cross-government report is equally clear that Copilot "showed limitations when dealing with complex data, especially for work requiring nuance and context". If most of your staff spend their day writing and reading, the case is stronger. If they spend it in spreadsheets, it is weaker than the survey numbers suggest.

What none of the trials could tell you is where the saved time went. The cross-government report says so directly: it "was not possible to identify how time saved was spent". Twenty minutes saved on an email that nobody spends on anything else is not a saving. It is a slightly shorter day.

Why your rollout proposal will not mention this

The people proposing a company-wide rollout are usually paid on seats. That is not a criticism; it is how licensing works. But it means the business case you are shown will lead with the most flattering public number, and 26 minutes a day is the most flattering one available.

There is also a quieter cost. Licences bought for everyone tend to go unused by a large share of the people who receive them, and each idle seat is a few hundred pounds a year. DBT's usage figures show what that looks like even in an organised pilot with volunteers. We made a related point about whether you need AI for the task at all: the expensive mistake is rarely the tool, it is buying it ahead of a specific job for it to do.

What to do before you sign

You do not need a three-month government evaluation. You need a few weeks and some discipline.

• Pick three tasks you can time. Real ones your staff do weekly: a monthly report, a supplier summary, a spreadsheet reconciliation. Time them now, without Copilot.
• Give licences to a small group, not a department. Twenty to fifty people across different roles, including a couple of sceptics.
• Time the same tasks again, and mark the output. Have somebody who did not do the work check the quality, as DBT did. Faster and wrong is not a saving.
• Check the usage report after a month. Microsoft's admin centre shows who has actually used their licence. Seats nobody opened are your answer for those roles.
• Ask what the time goes to. If a team saves three hours a week, what will it do with them? If nobody can say, budget for zero.

Then buy for the roles where the numbers held up, and not for the rest. That is the same discipline we argued for in testing any vendor's accuracy claims: measure on your work, not on somebody else's survey.

Where to start

If the renewal is close and you want a second opinion on the proposal before you commit, that is a day's work. Our sanity check costs £1,250: we read the business case, look at what your teams actually do, and give you a written answer on which roles are worth licensing and how to test the rest. Sometimes that answer is "buy it". Sometimes it is "buy forty, not four hundred".

Copilot may well pay for itself in your organisation. The point is to know which parts before you pay for all of it.