Microsoft 365 Copilot Cut Email Time in a 66-Firm Trial

What did AI actually solve?

AI did not solve knowledge work in general. It solved a narrower and more believable problem inside it: how much of the week email can consume. In this large randomized field experiment, integrated generative AI cut Outlook time meaningfully without obvious changes in task mix. That is a credible workflow gain, but it is still not proof of better work or a redesigned job.

What changed

Many work-AI studies ask whether people can write faster or complete a flagship task better. This paper asked a more ordinary question that still shapes the workday: can an integrated AI tool shrink the time knowledge workers lose to email?

Microsoft partnered with 66 large firms during the pilot rollout of Microsoft 365 Copilot and ran a six-month randomized field experiment. Firms recruited workers who already depended on Microsoft tools for email, meetings, and writing, then randomly assigned Copilot access within those groups. The analyzed sample included 7,137 knowledge workers, with 3,684 assigned to treatment.

The main result was concrete. Intention-to-treat estimates show that workers given Copilot spent 1.4 fewer hours per week in Outlook, a 12% reduction relative to the pre-period mean. Instrumenting actual use with randomized assignment, the paper estimates that workers induced to use Copilot spent about 2 fewer hours on email each week, a 17% reduction.

The restraint is that the rest of the job barely moved. Treated and control workers replied to the same number of email threads, attended the same number of Teams meetings, and completed the same number of Word documents. The paper also reports about 0.25 fewer hours per week of out-of-hours activity in the observed apps, but no broader reorganization of task mix.

What this could change for you

For office workers, the practical promise is not that AI will finish the job for you. It is that a tool embedded inside the software you already use may compress one of the most repetitive parts of it: reading, searching, and drafting around email.

If that kind of gain holds up in other settings, the most noticeable benefit may be a calmer workday rather than a dramatic productivity headline. Saving time on inbox work can matter even when the number of meetings, documents, and replies stays the same, because email is often the friction that spills into evenings.

For managers, the paper is a warning against evaluating AI only through spectacular output tasks. Summaries, inbox triage, and meeting follow-up may create value precisely because they remove low-grade coordination overhead. But the study also shows that adoption is not automatic: usage peaked early, then stabilized far below every treated worker using the tool every week.

Try it yourself

Check one email summary

Use an Outlook account with Copilot summarization enabled. Availability depends on your account, plan, and organization’s settings.

In Outlook, open an email conversation you are allowed to use with Copilot.
Choose Summary by Copilot or Summarize at the top of the thread. Microsoft’s guide covers the differences between Outlook versions.
Compare the summary with the original messages. Check who owns each action, any deadlines, and anything it missed.

What to watch for: Count the time spent checking and correcting the summary, too. This small trial cannot predict your weekly savings or reproduce the 66-firm experiment.

See Microsoft’s Outlook guide ↗

Suggested activity. Official guidance checked September 13, 2026.

What it does not prove

This was a study of large, early-adopting firms using Microsoft's integrated product during a limited-license pilot. It does not prove the same effect would transfer unchanged to smaller firms, non-Microsoft workplaces, hourly frontline work, or jobs that are not email-heavy.

The paper measures telemetry in Outlook, Teams, and Word rather than direct output quality, promotion, pay, or firm performance. It does not show that workers made better decisions, wrote better documents, or turned the saved time into higher-value work.

Because treated workers were free to use or ignore Copilot, the larger two-hour estimate is a local effect for workers induced to use the tool by random access, not a guarantee that every broad rollout will save the same amount of time.

Primary research