Microsoft 365 Copilot rollout plan: what the training needs
Use findings from UK government Copilot trials to plan practice time, sessions by role, clear usage rules and checks on the work.
Before booking Copilot training, agree with line managers which tasks their teams will practise and when people will have time to try them. Decide which material may go into the tool and who checks the resulting work.
Three evaluations, and what each found about learning
The evidence comes from a cross-government Copilot experiment with about 20,000 staff and from two departments that also evaluated their own trials.
The Government Digital Service ran the experiment from 30 September to 31 December 2024, and the Department for Work and Pensions (DWP) and the Department for Business and Trade (DBT) took part in it.
Their methods differ, and DBT says its findings are not comparable with other departments'.
| Evaluation | Who took part | How staff learned | Training finding |
|---|---|---|---|
| Government Digital Service, cross-government experiment, 2024 | Who took part About 20,000 staff in 12 organisations; 7,115 survey responses | How staff learned Resources from the centre, with user support left to each organisation; short training videos for staff short of time | Training finding The professions and departments least familiar and confident with AI saw lower benefits |
| DWP, trial from October 2024 to March 2025 | Who took part 3,549 central office staff who volunteered or were nominated; 1,716 user survey responses | How staff learned 89 percent of 1,713 surveyed users explored it on their own and 61 percent learned from peers; most interviewees did not remember a formal session | Training finding Users asked for sessions by role, shorter sessions and a prompt sheet for their department |
| DBT, pilot from October to December 2024 | Who took part 1,000 licences, about 70 percent to volunteers and 30 percent at random; 300 diary respondents | How staff learned Departmental and Microsoft sessions, and self-led learning that staff recorded | Training finding Two hours or more of self-led training went with higher satisfaction; formal sessions were seen as introductory |
Protect time to practise on their own work
Put a few protected hours for each person into the plan, to practise on their own tasks.
In DBT's diary study of 300 licence holders, respondents who had spent at least two hours on independent training and upskilling were more likely to be satisfied and to recommend the tool. For departmental training, only the group reporting at least five hours showed a statistically significant association with higher satisfaction. In interviews, staff described the departmental and Microsoft sessions as introductory or high level and less relevant to their roles, and found their own experimenting more useful for seeing where Copilot could help them.
Those staff chose their own hours and may have been keener from the start, so the hours alone do not prove the effect. Time was also short: in the cross-government experiment, some staff said their workload kept them from the upskilling, and the organisers added short training videos.
Book the time in the plan, and agree with each team which of its own documents, emails and meetings people may use for practice in the tool. The same rule holds for what the course itself should cover: start from the tasks in people's own jobs.
Run short sessions for each role
Follow an introduction to the basics with short sessions by role and a prompt sheet for each role's tasks.
DWP's licences went to volunteers and nominated staff, first come, first served. Its users asked for sessions shared with people in similar roles, shorter sessions on one feature at a time, and a cheat sheet with prompts tailored to DWP.
DBT's interviewees found Copilot best suited to roles with heavy administrative work or large volumes of documents, and less suited to legal and policy roles that depend on context, nuance and attention to detail.
Microsoft's own enablement material now includes prompt packs by function, for sales, marketing, human resources, finance, legal and others. Use them as a first draft, and replace the examples with your own teams' tasks, documents and rules.
In AI training, participants practise in the tools the company has approved and learn where a person must decide. I tailor the content to your company and to the department taking part.
Teach the rules and the check in the course
Teach which data may go into which Copilot product, and how to check what it produces, in the same sessions.
At DBT, knowledge of the acceptable use policy varied widely, and some staff had not read the security guidance. Among those who had tried to familiarise themselves with the guidance, there was confusion between Microsoft 365 Copilot and Copilot Chat, both available to them, and many who were unsure of the policies said they had been overly cautious, which limited their use.
Copilot works with each user's own access to files. The Government Digital Service warns that this may surface files a user can open but should not, so review permissions before the rollout.
Of the 300 DBT diary respondents, 22 percent said they had found false or misleading information presented as fact, 43 percent said they had not, 11 percent were unsure, 22 percent did not answer and 3 percent had not used Copilot during the trial. The reported percentages are rounded and describe what respondents noticed, not the error rate across all outputs. Some interviewees said they might not review an output closely if they were not sharing it. Several DWP users noted that vague prompts could lead to hallucinations or irrelevant results.
In each session, have participants check one output against its source document, and agree who answers for the result. Then ask in the first survey whether people review what they keep.
Give line managers and early users a part
Brief line managers before their teams start, and let early users show colleagues what worked.
At DWP, several users named their manager as the reason they applied for a licence, and early adopters became informal champions who showed colleagues the tool live. At DBT, the attitudes of a person's team and line manager appeared to shape how much they used Copilot, and staff without a licence said evidence of successful uses would inform whether they took it up.
Microsoft's kit includes onboarding templates for managers as well as users. Add a short session where managers try the tool on their own team's work, and name one early user in each team to show the others a task that worked.
Decide what counts as an active user
Define the reporting period and what counts as use before you report adoption.
In one pilot, DBT's dashboard showed that on average 64 percent of licence holders used Copilot in a given week, and about 30 percent on a given working day. Decide which reporting period you need and keep the definition consistent as the rollout continues.
Report use by application too. In the Government Digital Service's usage data for 14,500 users, adoption in Excel and PowerPoint peaked at 23 and 24 percent, respectively. Its definition counted at least one interaction in the application in the previous 30 days. For a team that works mainly in spreadsheets, ask which tasks people tried and what happened before deciding whether another training session would help.
Sources
- Microsoft 365 Copilot Experiment: Cross-Government Findings Report. Government Digital Service, 2 June 2025.
- An Evaluation of DWP's Microsoft 365 Copilot Trial. Department for Work and Pensions, 29 January 2026.
- The Evaluation of the M365 Copilot Pilot in the Department for Business and Trade. Department for Business and Trade, August 2025.
- Microsoft Copilot user engagement tools and templates. Microsoft, undated.
Facts last checked on 27 October 2026.