Reference resource · Microsoft 365 Copilot

Measuring real Microsoft 365 Copilot adoption: a five-level model

Most Copilot dashboards mix up five different things: having access, being trained, using, adopting, and getting a business result. This model separates them, with for each level the question to ask, the indicator, the Microsoft 365 source and the trap to avoid.

By Antoine Billotte, founder of Tomorrow Solutions · Updated 30 October 2026

Diagnose your Copilot adoption

Licences everywhere, no one knows who uses it

This spring, an international mid-sized industrial group walked me through its Copilot rollout. Almost every employee had a licence. Leadership had set a 100% training target. One number circulated at the executive committee: the completion rate of an e-learning module, which was very low.

No one could say how many people really used Copilot, or for what. Meanwhile, the licence bill arrived every month. This is the situation I meet most often, and its cause almost always lies in how adoption is measured.

Three indicators that reassure for the wrong reasons

These numbers are accurate within their scope. The risk lies in using activity signals to answer a question about value.

Licences assigned

It measures a purchase. It tells you how many people can use Copilot, not how many do.

"Active" users

In the Microsoft 365 admin center usage report, a user is active if they deliberately triggered a Copilot feature during the selected period (7, 28, 90 or 180 days). Opening the Copilot pane does not count; submitting a prompt does. A single prompt in four weeks is enough.

Assisted hours and assisted value

The Microsoft Copilot Dashboard shows "Copilot assisted hours". Microsoft itself describes them as an estimate based on actions and research multipliers. The assisted value multiplies these hours by an hourly rate set to 72 dollars by default.

The model: five levels, five questions

Each level answers a different question. You do not move to the next one until the previous one is measured.

  1. 01

    Access · Who can use Copilot?

    Indicator
    Licences assigned versus target population, by department and job family.
    Source
    Copilot usage report in the Microsoft 365 admin center ("enabled users").
    Trap
    Reasoning at group level. A 90% overall rate can hide a whole department without licences, or licences given to people with no use for them.
  2. 02

    Training · Who knows how to use it in their own job?

    Indicator
    Share of users trained on use cases from their own job, not just registered for a generic tool overview.
    Source
    HR or learning systems, plus one question at the end of each session: "Which task from your week will you hand to Copilot?"
    Trap
    Mistaking attendance for skill. Generic training, in a language that is not the team’s, with no time freed up by the manager, produces registrations and little use.
  3. 03

    Usage · Who has tried it, and how often?

    Indicator
    28-day active user rate (active divided by enabled), average prompts per active user, number of active days.
    Source
    Copilot usage report in the admin center, with the per-app view (Outlook, Teams, Word, Excel, PowerPoint, Copilot Chat).
    Trap
    Stopping here. This level measures exploration. It is the right indicator for the first weeks, not for judging a programme after six months.
  4. 04

    Adoption · Who uses it regularly?

    Indicator
    Share of regular users: in the Copilot Dashboard, "habitual" users (1 to 14 actions per week on average) and "power" users (15 or more) have used Copilot in at least 9 of the past 12 weeks. Add returning users, broken down by job family.
    Source
    Microsoft Copilot Dashboard in Viva Insights. Viewing it requires neither a paid Viva Insights licence nor a Copilot licence; benchmarks and group-level data open from 50 licences.
    Trap
    Reading an average. A legal team at 70% regular users and a sales team at 10% give an average of 40% that describes neither, and they call for two different action plans.
  5. 05

    Impact · What has changed in the work, and who is accountable?

    Indicator
    One or two business indicators per use case, measured before and after: request processing time, review cycles before sign-off, preparation time for a recurring meeting, quality perceived by the internal client.
    Source
    Business systems (request management, CRM, legal tools), plus the Copilot Dashboard sentiment survey on work quality, mental effort on repetitive tasks, speed and productivity.
    Trap
    Measuring without an owner. An impact indicator only has value if a business leader owns it, follows it and decides what comes next.

Set up the measurement in four weeks

Week 1

Pull the baseline

Ask IT for the Copilot usage report over 28 and 90 days, and open the Copilot Dashboard. It is free and quick, and it turns an impression into a fact.

Week 2

Pick two or three use cases per job family

Start from what takes time every week in the team. For each case, name a business sponsor and measure the starting point.

Week 3

Set the ritual

A short monthly review per job family: levels 3 and 4 to steer support, level 5 to decide. Each team’s champions bring the cases that work.

Week 4

Write the decision rule

For each pilot, a criterion written in advance to scale it, adjust it or stop it. This is what prevents pilots that run on without a conclusion.

Four mistakes to avoid

  1. Counting training registrations as an adoption indicator.
  2. Presenting assisted hours as a measured return on investment.
  3. Aggregating job families into a single rate that hides the teams at zero.
  4. Confusing a target with a result. An adoption target set at launch is not a result until it has been measured.

Leaders are starting to equip themselves for this question: according to KPMG France’s Trends of AI 2026 study of 356 decision-makers, two organisations in three now say they can measure the return on investment of AI, against one in three a year earlier. The remaining step is to check that this measurement covers the work itself, beyond activity in the tool.

Where to start

If you do only one thing this month: ask for the 90-day Copilot usage extract, broken down by department, and look at the share of regular users job family by job family. This single reading is usually enough to know where to focus support.

Frequently Asked Questions

By separating five levels: access (licences assigned), training (people trained on use cases from their job), usage (28-day active users), adoption (regular users, at least 9 weeks out of 12 in the Copilot Dashboard) and impact (a business indicator measured before and after, owned by a sponsor). Each level answers a different question.

Sources

Sources consulted on 30 October 2026.

Know what your licences are really producing.

I build these measurement systems with business leaders and IT, from the first extract to the monthly review.

Let’s talk about your rollout

Confidential · No commitment · FR / EN