Adoption

How to measure Copilot adoption without fooling yourself

By Matteo Garoli 6 min read

A white outline line chart with a rising zigzag line and an arrow on a frosted glass tile, floating over soft amber and ice blue light on a near-black background.

The two numbers most Copilot adoption reports start with are licences assigned and active users. Both are easy to get, and both can reassure you when they shouldn’t. In the Copilot Dashboard’s active-user figure and in the admin center usage report, a person counts as active after one deliberate action in the period, so a single question asked in four weeks scores the same as daily use.

I work as a Microsoft 365 Copilot Cloud Solution Architect (Microsoft’s documentation now calls the product Microsoft Copilot). This year I’ve run Copilot adoption reviews with large organisations across Europe.

Why isn’t the number of active Copilot users enough?

Because “active” is a low bar. In the Copilot Dashboard in Viva Insights, an active Copilot user is someone who performed at least one Copilot activity in the previous 28 days. The admin center usage report sets the bar in the same place.

Opening the Copilot pane in Word doesn’t count, but submitting a prompt does. A team that asked one question last month and a team that opens Copilot every morning can show the same active users rate. The rate tells you whether people tried Copilot, and nothing about whether a habit formed.

Self-reported pictures can mislead too. At one transport operator, the answers written down beforehand suggested low adoption, because whoever filled them in had agents in mind. In conversation it turned out chat and the Office apps were in everyday use; the real gap was agents, which nobody owned. I’ve seen a mismatch between the written picture and the spoken one in four reviews, so I never rely on a single source.

What should you measure instead of licences and active users?

Four things: how often people use Copilot, how many apps they use it in, whether they come back, and what they use it for.

For frequency, the Copilot Dashboard sorts licensed users into usage levels over a rolling 12 weeks. A habitual user averages between 1 and 14 Copilot actions a week and was active in at least 9 of the 12 weeks. A power user does the same with an average of 15 or more. A novice user averages at least one action but was active in fewer than 9 weeks. Everyone else, including people who haven’t used Copilot yet, is “other”.

Days of use give a second view of frequency. The AI adoption score in Adoption Score counts, for each licensed user, the days out of the past 28 on which they used Copilot, divides that by 12, and averages across licensed users. A score of 100 means an average of at least 12 of 28 days, which is three a week and Microsoft’s own target; a score of 50 means 6 of 28 days. Days of use show how close you are to that target, while usage levels, which weigh actions per week, show which groups are behind. I track the first at tenant level and use the second to decide where to train.

For breadth, the adoption view splits active users by app: Teams, Outlook, Word, Excel, PowerPoint and Copilot Chat. A healthy overall figure can hide an app nobody touches, or a person who only ever summarises meetings.

Returning users are people with at least one action in both the current and the preceding period, which the dashboard shows monthly or weekly. If active users climb while returning users stay flat, you are collecting first tries rather than habits.

For the task itself, the dashboard sorts Copilot Chat prompts into “Ask and find”, “Catch up”, “Draft and brainstorm” and other. That is as close as the reports get to the task. For anything finer I talk to a handful of people and ask what they used Copilot for last week, what they tried once and dropped, and what they would show a colleague. A few of those conversations are enough to show whether the numbers match daily work. Adoption Score also suggests a four-statement survey on quality of work, mental effort, speed and productivity, which a Global administrator can upload into the Copilot Dashboard.

What do the Copilot Dashboard and the admin center reports show?

The admin center usage report is written for IT admins and answers who is licensed and who used Copilot. The Copilot Dashboard is written for leaders and managers and adds usage levels, returning users, group breakdowns, impact estimates and survey results.

The admin center report shows Enabled Users, Active Users and an Active users rate (active divided by enabled), with adoption by app, over 7, 28, 90 or 180 days. Microsoft’s own pages differ on 28 versus 30 days, so check the filter in your tenant. A table lists each user’s last activity date per app, with names hidden by default, and the data arrives a couple of days late.

The dashboard looks at the previous 28 days, with a delay of up to six days. Any tenant with at least one Copilot licence gets the adoption and impact pages. With at least 50 Copilot licences or 50 Viva Insights licences you also get benchmarks, trendlines, the manager view and survey data. Microsoft says both use the same data, so differences come from timeframes, delays and privacy thresholds.

Where do Copilot usage numbers mislead you?

They count deliberate actions by licensed people, some headline figures are estimates, and Microsoft’s own notes list recent changes to how some figures are calculated.

Copilot assisted hours is the clearest example. It multiplies counted actions by a fixed assistance factor: 6 minutes for most search, summary and creation actions, and the length of the meeting for meeting summaries. Microsoft calls the result a general estimate. Copilot assisted value then multiplies those hours by an hourly rate, which defaults to $72 and is based on US Bureau of Labor Statistics data. Nothing in that calculation looks at output. I measure return on one process, before and after, using whatever that team delivers.

Groups below the minimum group size disappear from the group view. The satisfaction rate appears only with at least 30 responses from at least five users in 28 days. Microsoft merged Copilot Chat (work) and Copilot Chat (web) into one Copilot Chat metric, and a known issue understated email draft and email summary figures, along with assisted hours and value, between 1 June 2025 and 18 February 2026. Check those notes before comparing this quarter with last year. Microsoft also says Copilot is unlikely to be solely responsible for any change you see, because seasonality, role shifts and organisational changes move the same numbers.

How do you turn the numbers into the next training session?

Find the group where the numbers and the people disagree, learn which tasks that team should be doing with Copilot, and build the session around their own work.

  1. Open usage levels by group and by app in the dashboard. Look for a department with licences and mostly novice or other users.
  2. Put satisfaction next to usage. Microsoft’s guidance is that high usage with low satisfaction may mean people are trying hard and struggling.
  3. Ask a handful of people in the group the three questions above and note the tasks that come up.
  4. Build the session around those tasks, using the team’s own documents and meetings. How I design that training is in my article on why Copilot licences are not adoption.
  5. Wait before measuring. Returning users compare 28-day periods and usage levels use a rolling 12 weeks, so judge returning users after a month and usage levels after about three months.

If you have your own numbers in front of you and can’t tell what they mean, write to me.

Sources

If you are working on Copilot adoption, a training or a talk, or you are starting out in this world, write to me.