Short answer: low Copilot adoption is rarely a software fault. It is usually a rollout fault, and a measured 90-day plan is a practical way to tackle it. Count who is actually active before you do anything else. Then give each team one named job for the tool, fix the permissions that make people (and security) nervous, train through real tasks, run a small group of champions, and take the seats back from anyone still idle at day 90. Public evidence supports this. In a UK Department for Business and Trade pilot of 1,000 licences, mean daily active use was only 21%, even though about 70% of the users had volunteered. A bought licence is not an adopted tool.
This guide is for the person who signed off the seats and now has a usage report full of grey bars. It uses Microsoft 365 Copilot as the example because it is a common per-seat purchase, but the plan applies for any per-seat AI assistant. Prices are US list prices as of 20 November 2026, and the research is cited as third-party studies. Nothing here is legal advice.
Why do people stop using Copilot after the launch email?

A typical rollout goes like this. Licences are assigned, an all-staff email goes out and a few curious people try it. After two weeks nobody owns the number, so it drifts. Each cause below has a fix you control.
| Cause | What it looks like | Fix |
|---|---|---|
| No named job for the tool | What it looks like: People open it, ask something vague, get an average answer and leave | Fix: Give each team two or three specific tasks to try first (see the 90-day plan) |
| Seats given to the wrong people | What it looks like: Licences assigned by seniority or by who asked first, not by work that suits the tool | Fix: Match seats to roles with heavy email, meeting, document and spreadsheet work |
| Fear about what Copilot can see | What it looks like: Staff and security teams worry it will surface files people should not read | Fix: Review sharing and permissions before wide rollout |
| Generic training | What it looks like: A one-hour webinar that shows features, not the user's own work | Fix: Short, task-based practice inside each team |
| Nobody measures | What it looks like: No baseline, no weekly number, no owner | Fix: Pull the usage report weekly and name an owner |
| Weak fit for some tasks | What it looks like: People try a task the tool does badly and write it off | Fix: Steer people to tasks where it helps and say plainly where it does not |
The UK evidence on the last two rows is useful. The Department for Business and Trade (DBT) evaluation found that some tasks, such as scheduling and image creation, took longer with Copilot. It also found that self-led training was more effective than formal sessions. That does not mean training is pointless. It means a lecture is the wrong shape for it.
What does an unused Copilot seat actually cost?
Start with the seat price. These are public US list prices for Microsoft 365 Copilot, ex-tax, as of 20 November 2026. Local pricing varies, and a Copilot licence also needs a separate qualifying Microsoft 365 plan.
| Product | List price (US, ex-tax) | Notes |
|---|---|---|
| Microsoft 365 Copilot | List price (US, ex-tax): $30.00 per user per month, paid yearly; $31.50 paid monthly on an annual commitment | Notes: Needs a qualifying Microsoft 365 plan |
| Microsoft 365 Copilot Business | List price (US, ex-tax): Shown as $18.00 per user per month, paid yearly, against an original $21.00 | Notes: For plans of up to 300 users. The discount runs 1 July to 31 December 2026, needs an annual commitment and applies to the first year only |
| Copilot Chat (web-grounded) | List price (US, ex-tax): No additional cost with eligible Microsoft 365 subscriptions | Notes: The paid licence adds grounding in your Microsoft Graph work data |
Some advanced features are also billed by usage ("Copilot Credits") on top of the seat price. Check which features you have switched on.
Here is simple arithmetic, not a finding. A 200-seat purchase at $30 a user a month on yearly billing is $72,000 a year. Each seat is $360 a year. If 40 of those seats have not been used in 28 days, that is $14,400 a year paid for nothing, before anyone counts the time spent on a failed rollout. Replace the 200 and the 40 with your own usage figures.
Microsoft publishes a Forrester Total Economic Impact study showing a return of over 100% with payback in 10 months. It is commissioned by Microsoft and models a composite organisation over three years. Treat it as a vendor claim to test against your own usage.
How do I measure Copilot adoption properly before changing anything?
Do this first, in week one. A plan with no baseline cannot show whether it worked.
The Microsoft 365 admin center has a Copilot usage report under Reports, then Usage. It can be viewed over the last 7, 28, 90 or 180 days. Its main numbers are:
- Enabled users: unique users with a Copilot licence in the period.
- Active users: enabled users who tried a user-initiated Copilot feature in one or more Microsoft 365 apps.
- Active users rate: active users divided by enabled users.
- Per-user table: prompts submitted, active days and the last activity date for each licensed user. It covers anyone licensed at any point in the past 180 days, including people who never used it.
Three details matter. First, "active" means an intentional action. Opening the Copilot pane does not count. Submitting a prompt does. Second, user names are hidden in usage reports by default, so an admin has to change that setting before you can see who is idle. Check with your privacy or HR owner first. Third, Microsoft says the Purview audit log is not meant for usage reporting, and totals built from it may not match the official report. Use the usage report or the Copilot Dashboard in Viva Insights for adoption numbers.
Export the per-user table to CSV on day one and keep it. Then record four numbers: licences bought, active users rate over 28 days, active days per user, and the number of licensed people with no activity at all. Those are your baseline.
What does a 90-day Copilot adoption plan look like?
This plan is a working structure, not a researched formula. Adjust the lengths to your size. It suits a company of a few hundred people with Microsoft 365 already in place.
| Days | Focus | What you do | Evidence you keep |
|---|---|---|---|
| 1 to 14 | Focus: Baseline and safety | What you do: Export usage, name an owner, list teams by licence count, review sharing on the sites and libraries Copilot will read, set a retention policy for Copilot interactions if your policy needs one | Evidence you keep: Baseline CSV, permission review notes, written usage guidance |
| 15 to 45 | Focus: Pilot in 2 or 3 teams | What you do: Pick teams with heavy email, meeting and document work. Give each two or three named tasks. Appoint champions. Hold short practice sessions on real work | Evidence you keep: Weekly active rate per team, task list, champion notes |
| 46 to 75 | Focus: Expand what worked | What you do: Roll the tasks that worked to the next teams. Retire the ones that did not. Publish a short prompt library written by the pilot teams | Evidence you keep: Updated task list, prompt library, weekly active rate |
| 76 to 90 | Focus: Decide on seats | What you do: Review each licence. Reassign seats with no activity in 28 days. Decide whether to renew, shrink or change tool | Evidence you keep: Seat decision record, renewal paper |
How do I stop Copilot surfacing files people should not see?
Microsoft states that Copilot only surfaces organisational data to which each user has at least view permission. That is a safe design, but it means Copilot follows whatever your SharePoint and OneDrive permissions already allow. A messy permission set therefore becomes a Copilot exposure. A folder shared with "everyone" years ago is now one prompt away.
Before you widen the rollout, run through the sharing on HR, finance, legal and board libraries, remove stale "everyone" and broad group access, and check guest access. Microsoft's Purview Data Security Posture Management for AI is designed to help discover oversharing risks. Fixing this early also answers the security team's first objection.
How do you pick the right tasks for each team?
Do not ask teams "how could you use AI?". Ask what they do every week that is repetitive, text-heavy or slow to start. Starting points include summarising a long email thread, drafting a first reply and turning meeting notes into actions. Each team should leave with two or three tasks in their own words. If you are weighing many ideas across the company, our use case prioritisation matrix shows how to rank them.
Ask pilot users how long the task took before, whether they trusted the result and whether they had to fix it. Some tasks got slower in the DBT report. Drop those from the list rather than push people to persist.
How do you run an AI champions programme?

A champions programme turns a few enthusiastic people into the help desk your IT team has no time to be. It directly addresses the "nobody owns this" cause above. It works only if the champions have time, a clear job and a number to move.
How do I pick champions?
- Choose people who already use the tool and who colleagues trust, not only the most senior or the most technical.
- Take at least one from every pilot team, so help is a desk away.
- Ask their manager to protect a small, fixed amount of time each week. Without it the programme dies in the first busy month.
- Include a sceptic or two, who will find weak spots early.
What do champions actually do?
- Run a weekly 30-minute open session on real work, not a presentation.
- Collect the prompts and workflows that save time and add them to the shared library.
- Log tasks where the tool failed, so the list stays honest.
- Help new users through their first two weeks.
- Tell the programme owner what the usage data cannot, such as why a team has gone quiet.
How do I measure a champions programme?
Use plain numbers you can pull without extra tooling: active users rate in the champion's team against teams without one, active days per user, the number of tasks in the shared library, and the count of idle licences at day 90. Take care when you read the results. Teams with champions may have been chosen because they were keen, so a higher rate is a hint, not proof.
There is no public evidence for an ideal champion-to-staff ratio, so we do not quote one. Start with one per team of about 15 to 25 people and adjust when you see whether they can keep up.
What kind of training works for Copilot?
Short, task-based and self-paced beats one big session. The DBT evaluation found self-led training more effective than formal sessions. Use that. A practical pattern is a short guided start on the team's own work, a written card per task with an example prompt, and a champion on hand.
Whatever you train, include three rules every user needs: check outputs before sending them on, do not paste information into unapproved tools, and report anything odd. A one-page usage policy is enough. If staff are already using other AI tools that you have not approved, a shadow AI audit shows which and why. The reason is often that the approved tool did not suit the job.
What results should I expect, and how far can I trust the studies?
Public evidence exists, and it is more cautious than vendor material.
| Study | What it covered | What it found |
|---|---|---|
| UK DBT Copilot pilot (report published 2025) | What it covered: 1,000 licences, October to December 2024. About 70% volunteers, 30% randomly assigned | What it found: Mean daily active users 21% (30% excluding weekends and bank holidays). Mean weekly active users 64%. About 1.14 Copilot actions per user per day. Satisfaction 72%. No evidence that time savings led to improved productivity |
| UK cross-government experiment (published 2 June 2025) | What it covered: 20,000 government employees in 12 organisations, 30 September to 31 December 2024 | What it found: Users reported saving an average of 26 minutes a day. Average satisfaction 7.7 out of 10. About 80% adoption was maintained. 82% would not want to go back |
Read the two together. The 26 minutes a day is self-reported, so say "users reported", not "Copilot saves". The DBT team found no evidence of measurable productivity gains. Both studies ran in late 2024 with public-sector staff, so neither proves your result. They suggest a range: many licensed people use the tool weekly, fewer use it daily, and users tend to like it. Set targets from your own baseline, not a headline.
When should I take seats back, or change tool?
Set the rule at the start and tell people, so reclaiming a seat is not a surprise. A reasonable rule is that any user with no activity in 28 days after day 60 loses the seat, with a simple route to ask for it back. Then compare the cost per active user with what the seat is worth to that team.
Copilot is not the only option, and your answer may be a mix. Copilot Chat is included at no extra cost with eligible Microsoft 365 subscriptions, so some people may be fine without a paid seat. Other tools price differently: Claude Team lists at $20 per standard seat per month billed annually ($25 monthly), Claude Team Premium at $100 annually ($125 monthly), and Claude Enterprise at $20 per seat per month billed annually with a 20-seat minimum and usage billed separately at API rates (list prices at claude.com/pricing, as of 20 November 2026). ChatGPT Business starts at two users, billed monthly or annually. See per-user pricing on OpenAI's site. ChatGPT Enterprise has no public price.
Check the data terms before moving. Microsoft says prompts, responses and data accessed through Microsoft Graph are not used to train its foundation models. OpenAI says it does not train on business data by default, and Anthropic says the same for its commercial products. Do not assume any of them keeps data in India. Microsoft says customers outside the EU may have queries processed in the US, EU or other regions. OpenAI can store eligible Enterprise content at rest in India, but model inference on it does not currently run in India. We make no residency claim for Claude. For a detailed comparison for regulated firms, see ChatGPT Enterprise vs Azure OpenAI vs self-hosted, and for personal data in these tools see DPDP and ChatGPT or other AI tools.
Where can Mono help with an AI rollout?
Our AI adoption work covers discovery with each team, tool choice, pilots, training and a usage policy. The plan above is also something your own team can run without us.
Sources: Microsoft, Microsoft 365 Copilot enterprise and business pricing pages (microsoft.com), checked 20 November 2026. Microsoft Learn, "Data, Privacy, and Security for Microsoft 365 Copilot" (updated 4 November 2026) and "Microsoft Copilot usage report" (updated 9 November 2026). UK Department for Business and Trade, "Evaluation of the M365 Copilot pilot in the Department for Business and Trade" (gov.uk, August 2025). UK Government Digital Service, "Microsoft 365 Copilot experiment: cross-government findings report" (gov.uk, 2 June 2025). OpenAI enterprise privacy page and data residency help article, November 2026. Claude pricing page and Anthropic privacy centre, November 2026. Forrester Total Economic Impact study commissioned by Microsoft (March 2025), as cited on Microsoft's site. Last reviewed 20 November 2026.
FAQ
Why is nobody using our Microsoft 365 Copilot licences?
Usually because the rollout gave people a licence but no specific job for it, no relevant training, no owner and no measurement. In a UK Department for Business and Trade pilot, mean daily active use was 21% even though about 70% of users had volunteered. Fix it by pulling the usage report, giving each team two or three named tasks, training on real work and reviewing every seat at day 90.
How do I check Copilot adoption in my organisation?
Use the Microsoft Copilot usage report in the Microsoft 365 admin center (Reports, then Usage). It shows enabled users, active users, the active users rate and a per-user table of prompts, active days and last activity date, over 7, 28, 90 or 180 days. User names are hidden by default and must be switched on by an admin. Microsoft advises against using audit log data for usage reporting.
How much does Microsoft 365 Copilot cost per user?
At US list prices checked on 20 November 2026, Microsoft 365 Copilot is $30.00 per user per month paid yearly, or $31.50 paid monthly on an annual commitment. It needs a qualifying Microsoft 365 plan. Copilot Business, for up to 300 users, was shown at $18.00 per user per month under a first-year discount running 1 July to 31 December 2026. Local pricing varies, and some features use usage-based billing.
What does an AI champions programme involve?
A small group of respected staff, at least one per pilot team, with protected weekly time. They run open practice sessions on real work, collect working prompts into a shared library, log tasks where the tool fails and help new users in their first two weeks. Measure it with active users rate, active days per user and idle licences at day 90, and treat team comparisons as hints rather than proof.
Does Copilot save time, according to public studies?
Evidence is mixed. In a UK cross-government experiment with 20,000 government employees, users reported saving an average of 26 minutes a day, but that is self-reported. The separate Department for Business and Trade pilot found no evidence that time savings had led to improved productivity, and some tasks took longer. Measure your own teams on named tasks rather than relying on either figure.
When should we take Copilot seats away from users?
Set a rule at the start, for example removing seats with no activity in 28 days after the 60-day mark, with an easy way to ask for the seat back. Review at day 90 with your usage data. Include Copilot Chat, which comes at no extra cost with eligible Microsoft 365 subscriptions, and other assistants in the comparison before you renew.
