Guide

Measuring Microsoft Copilot adoption and value

How to show what changed after a Microsoft 365 Copilot rollout or AI training programme: a four-level framework, the data sources in the Microsoft stack, and how to report without overstating.

By Josh Igoe, Microsoft Certified Trainer · Updated 14 October 2026

The question every sponsor asks

Sooner or later, someone asks whether the investment in Microsoft 365 Copilot, Microsoft Copilot Studio or Power Platform is paying off. Finance wants to know whether to renew licences. Leaders want to know whether to expand. L&D wants to show that training changed something beyond attendance.

The honest answer is usually ‘we are not sure’, because nobody agreed what to measure before the rollout began. This guide sets out a practical measurement framework for Microsoft Copilot adoption and AI training, the data sources available inside the Microsoft stack, and how to report results credibly without overstating them.

Five principles for credible measurement

1. Agree measures before you start. A baseline taken after training is not a baseline.
2. Measure a few things well. Three to six measures that people understand beat a dashboard of forty.
3. Separate activity from value. Usage tells you people are trying. It does not tell you their work improved.
4. Label self-reported data honestly. ‘Learners reported saving around an hour a week’ is credible. ‘Microsoft Copilot saved 10,000 hours’, extrapolated from a survey, is not.
5. Report what did not work. Weak adoption in one team is useful information. It tells you where to focus next.

A four-level framework

We use four levels, adapted from well-established training evaluation practice and applied to Microsoft AI adoption. Each level answers a different question.

Most organisations measure level 1 well and level 4 poorly. The aim is not perfect measurement at every level, but enough evidence at each level to tell a coherent story.

Data sources in the Microsoft stack

Microsoft 365 Copilot usage

The Microsoft 365 admin centre includes a Microsoft 365 Copilot usage report showing how many licensed users are active over a period and in which apps, such as Teams, Outlook, Word, Excel and PowerPoint. It is the simplest starting point for level 3. Depending on your licensing and configuration, the Copilot Dashboard in Viva Insights provides more detailed adoption and impact views. Check what is available in your tenant and who is permitted to see it, and be aware that some reports anonymise or aggregate data by default.

Copilot Studio analytics

Each Copilot Studio agent has analytics covering sessions, engagement, resolution, escalation and satisfaction where enabled. For agents, these are often the most direct measures of value: how many conversations were resolved without a person, and which topics fail most often.

Power Platform

The Power Platform admin centre provides analytics on apps and flows, and the Centre of Excellence (CoE) Starter Kit builds an inventory of makers, apps, flows and connectors. Flow run history shows how often an automation runs, which is the starting point for estimating time saved.

Surveys

A short survey before training and again four to eight weeks afterwards remains the best source for confidence, perceived quality and self-reported time saved. Keep it to five to eight questions and ask the same ones each time.

Process data

For specific automations and agents, the best evidence is in the process itself: ticket volumes, cycle times, rework rates or backlog. Agree with the process owner which figures to use and how to read them.

A survey you can reuse

These questions work before and after training. Use a consistent scale, such as 1 to 5.

• How confident are you using Microsoft 365 Copilot in your day-to-day work?
• In a typical week, how often do you use Microsoft Copilot? (Never / Once or twice / Several times / Daily)
• Which tasks do you use it for? (Choose from a list matched to your scenarios)
• When you use Microsoft Copilot, how often is the result useful with little or no editing?
• Roughly how much time does Microsoft Copilot save you in a typical week? (None / Under 30 minutes / 30 to 60 minutes / 1 to 2 hours / More than 2 hours)
• What stops you using it more? (Free text)
• After training only: what is one thing you now do differently?

The free-text answers are often the most useful part. They tell you where the barriers are: licences, data access, confidence or time.

Estimating value without overstating it

Time saved is the measure sponsors ask for most, and the easiest to overstate. A cautious approach:

1. Use banded self-reported time from the survey, and take the lower end of each band.
2. Apply it only to people who report using Microsoft Copilot regularly, not to everyone licensed.
3. Present it as a range and label it as self-reported.
4. Where possible, validate with a small number of task-level examples: ‘the monthly report now takes two hours rather than five, according to the team that produces it’.
5. Do not convert hours into cash savings unless finance agrees the method. Time saved is usually redeployed rather than removed.

For agents and automations, measure the process directly: conversations resolved, tickets avoided, approvals completed faster. These figures are more defensible than survey estimates because they come from system data.

What to measure for each type of programme

Microsoft Copilot rollouts (users)

• Active Microsoft Copilot users as a share of licensed users, by app and by team, before and after.
• Confidence and frequency of use from the survey.
• Self-reported time saved, banded and labelled.
• Champions active and team sessions held.

Agent builder programmes (builders)

• Builder confidence before and after.
• Agents designed, built and reaching production.
• For production agents: sessions, resolution rate, escalations and satisfaction.
• Governance in place: environments, data policies and publishing approvals.

AI automation programmes

• Makers trained, and still building after three and six months.
• Solutions in production, with owners and the processes they address.
• Hours saved or errors avoided, validated by the process owner.
• Certification results.

A one-page report for your sponsor

Keep the report short enough to be read. A structure that works:

1. What we set out to do: audience, scenarios and agreed measures, in three lines.
2. What we delivered: cohorts, attendance and completion.
3. What changed: the before-and-after figures for each agreed measure, with sources and labels.
4. What people told us: two or three representative quotes from learners (anonymised) and the main barriers.
5. What we recommend next: specific actions for teams with weak adoption, the next cohort or role group, and the next step up.

A version of this template is available in our resources.

Common pitfalls

• No baseline. Without a before, an after means little.
• Counting licences as adoption. Assigned is not active.
• Vanity totals. Large cumulative numbers (‘50,000 prompts’) say nothing about value.
• Extrapolating surveys. Multiplying a self-reported average across the whole organisation overstates the result.
• Measuring too early. Habits take weeks to form. Measure again four to eight weeks after training.
• Ignoring privacy. Agree with HR and your data protection lead how usage data will be used, and report at team level rather than about individuals.

An illustrative measurement plan

Here is how the framework comes together for a typical Microsoft Copilot rollout to three teams. This is an illustrative plan, not results from a client.

At scoping, the sponsor and adoption lead agree five measures: confidence using Microsoft Copilot; share of licensed users active each week; frequency of use; self-reported weekly time saved on the agreed scenarios; and one process measure per team, such as how long it takes to produce the monthly operations report.

Before training, the adoption lead captures four weeks of usage from the Microsoft 365 admin centre, sends the baseline survey and asks each team lead to record the current time taken for their process measure.

During training, the trainer records attendance and completion of lab tasks, and notes the barriers people raise: missing licences, data they cannot reach, uncertainty about what is allowed.

Four to eight weeks after training, the adoption lead repeats the survey, compares usage over an equivalent period and asks team leads to record the process measure again.

The report sets the before and after side by side, labels self-reported figures, quotes a few anonymised comments and recommends specific follow-up for any team where adoption lags, such as a champion clinic, a role-specific course or a fix to a data access problem.

Nothing here needs specialist tools. It needs measures agreed early and someone responsible for collecting them.

Frequently asked questions

Can we measure individuals' Microsoft Copilot use?

Technically some data exists at user level, but we recommend reporting at team or group level. Measuring individuals can undermine trust and raises data protection questions. Agree the approach with HR and your data protection lead first.

What is a reasonable adoption target?

It depends on roles, licences and how long Microsoft Copilot has been available, so set targets from your own baseline rather than published averages. A rising trend in regular use, sustained after training, matters more than any single figure.

Next step

Explore the current Microsoft Copilot training pathway for leaders, users and agent builders. Compare courses and prices

IGOE AI learning resource · Our guides are independent educational content, not a promise of course inclusions. See the current five-course training offer.