Copilot training for teams is structured teaching in how a group uses Microsoft 365 Copilot on its own work, delivered onsite, live online, self-paced or as a blend of those.
Choose the format by where your team's gap is. Live sessions suit hesitant staff and role-specific practice; self-paced suits confident people with protected time. In the Department for Business and Trade's Copilot pilot (October to December 2024), staff who did at least 2 hours of self-led learning were more likely to be satisfied.
That finding sits awkwardly for a company that sells live training, so it is worth saying plainly: ivee runs live and bespoke sessions, and for some teams self-paced is the better answer. This post compares four formats on the same criteria, says what a session should cover, and ends on who owns the follow-up, which matters at least as much as the format.
How do the four Copilot training formats compare?
Onsite and live online training win on role-specific practice and hesitant staff, self-paced wins on cost per head and flexibility, and a blend wins on mixed-ability teams. The table uses the same criteria for every format, so you can see where each one is weak as well as where it is strong.
Criterion | Onsite | Live online | Self-paced | Blended |
|---|---|---|---|---|
Best for | Hesitant staff and role-specific practice | Hybrid or multi-site teams doing similar tasks | Confident users with protected time | Mixed-ability teams |
Time per person (planning range) | Half a day to a day, plus travel | 2 to 3 hours, or two shorter sessions | 2 to 5 hours, spread over weeks | About 2 hours self-led, then one live session |
Main cost drivers | Trainer day rate, travel, room, the whole team out of the diary at once | Trainer day rate, smaller cohorts cost more per head | Any licence fee, plus hours nobody costs | The live fee, plus protected time for the self-led part |
Handles mixed ability | Only if the trainer can split the room by task | Breakout rooms help; quiet people stay quiet | Everyone sets their own pace; hesitant staff may never start | Best of the four: the confident skip ahead, the hesitant get a live start |
Where it falls over | A one-off day with no follow-up | Attention drifts, cameras go off | Nobody notices who has not started | Nobody owns the self-led half |
The cost row hides the biggest line, which is hours. Before you compare quotes, price the time: 96% of the 500 UK AI decision makers in ivee's August 2026 survey said they cannot show the numbers on their organisation's AI spend. ivee's free ROI calculator turns hours saved per person per week, team size and industry into an annual figure in about a minute, which gives you a number to set against any of the four.
What should a Copilot training session cover?
A Copilot session should cover the places people actually use Copilot, a reusable way of prompting it, and what to check before trusting an answer. Five things, in this order:
Copilot Chat. The starting point for everyone, and the place to practise prompting before touching a live document.
Copilot inside Word, Excel, Outlook and Teams. One exercise per application, built from the team's own documents rather than a demo file.
Structured, reusable prompts. A short template people can copy and adapt, such as the ones in best Copilot prompts for everyday work tasks.
Copilot Studio and Agent Builder. A short introduction for the two or three people who will build their own agents. Everyone else can skip it.
Checking output and handling data. What to verify before sending, and what must not go into a prompt under your own policy.
Skip the tour of every button. A good session spends most of its time on one real task per person, because a task from someone's own week is the kind most likely to be repeated on Monday. Microsoft's own Copilot skilling centre already covers the introductory material for free, which is a reason to spend paid time on the role-specific part instead.
Is onsite Copilot training worth it?
Onsite is worth it when the group is hesitant, the work is role-specific and the trainer can split the room by task. It is poor value for a confident team that mostly needs to hear what is new.
What you get from the room is observation. A trainer can see who is stuck at the first step, and a hesitant person may ask a question in person that they would never type into a chat. What you pay for is a whole team out of the diary at once, plus travel, and a day that has to be good because it will not be repeated.
A single day is also not a habit. In the Department for Business and Trade pilot, only staff who completed at least 5 hours of departmental training were significantly more satisfied with Copilot, and no other departmental training group differed significantly. Interviewees described departmental and Microsoft sessions as often introductory or high-level, and less relevant to their roles. If you book onsite, ask the provider to run the exercises on your own documents, and to show you the materials for one role before you sign.
When is live online Copilot training enough?
Live online is enough when your team is spread across sites or works hybrid, the tasks are broadly similar and you can keep groups to around a dozen. It gives you most of what onsite does at a lower cost in travel and diary time.
It also makes the two-session pattern easy: a first session on the basics, then a second two or three weeks later, once people have met real problems. That gap gives the second session real material to work on, which a single onsite day cannot offer.
The weakness is visibility. A trainer cannot see twelve screens, and the hesitant people are the ones who stay muted. Three fixes cost little: set one small task before the session, use breakout rooms by role rather than at random, and name one person per team whose job is to ask the question nobody else will.
When does self-paced Copilot training work better?
Self-paced works better when people already use Copilot, can protect a couple of hours, and need specifics more than motivation. The Department for Business and Trade evidence is the strongest case for it.
In that pilot, respondents who did at least 2 hours of independent training were more likely to be satisfied with Copilot and to recommend it, at a 99% confidence level, and interviewees found self-led experimentation more useful because it showed them where Copilot helped in their own role. Prior AI experience made no significant difference to satisfaction, which is a warning against sorting a team into confident and hesitant on a hunch.
There is a free baseline to beat. As of November 2026 Microsoft's skilling centre lists Copilot Academy learning paths, Virtual Training Days, instructor-led virtual events and a get started with Copilot learning path, and describes Copilot Chat as free and available with your Microsoft subscription. A paid self-paced course has to offer something those do not, usually role-specific content or a structure that gets people to finish.
Read the evidence with its limits in view. It is one department's pilot, built on self-reported diaries with a 32% response rate and no comparison group. ivee's general case for live cohorts, set out in live versus self-paced AI training, rests on different evidence, so treat the Copilot result as a caveat to that case rather than a replacement for it.
Who owns follow-up after a Copilot session?
One named person per team owns follow-up, and it should be the line manager or a champion they appoint, not the trainer and not IT. The evidence for that is softer than the evidence on training hours, but it points one way.
The Department for Business and Trade evaluation found that the attitudes of an individual's team and line manager appeared to influence how much, and how, they used Copilot. Usage in the pilot was light: the department's dashboard showed an average of 1.14 Copilot actions per user per working day, a figure that does not account for leave. A session does not fix that by itself.
Follow-up is four small jobs. Check active use six weeks on, using the Copilot Dashboard in Viva Insights if you have it. Keep one shared list of prompts that worked. Give each person one task from their own role to repeat weekly. Name where questions go. If people report that Copilot cannot find the right files, or finds ones it should not, that is a data problem the training will be blamed for, and the Copilot readiness checks are where to start.
Which Copilot training format should your team choose?
Choose by the gap, and measure the gap first, because the people who hold the budget often cannot say what their team can do. In what 500 UK AI decision makers told ivee in August 2026, 75% said they either have no idea, or that fewer than 1% of their workforce knows how to build and deploy an AI automation. That is a narrower skill than using Copilot, and the sample was drawn from organisations already building with AI, but it shows how rarely anyone has measured the spread.
Measure yours with two questions to each person: have you used Copilot this week, and what one task would you most like it to do? The answers sort a team faster than a skills test, and the second question gives you the exercises.
Mostly hesitant, role-specific work: onsite or live online, with one real task per person.
Mostly confident, short of time: self-paced, starting with Microsoft's free material, plus one live clinic for questions.
A mix, which is the usual case: blended. Ask everyone to do about 2 hours self-led first, then run one live session built around where people got stuck, then the six-week check.
Putting the self-led part first is our recommendation rather than a finding. It follows from the pilot evidence, and it means the paid live time goes on the problems people have actually met.
Where does each format fall over?
Each format fails in a predictable place, and the failure is rarely the content. Onsite fails when a single day is treated as the whole programme. Live online fails when the hesitant stay muted and nobody checks. Self-paced fails when nobody notices who has not started. Blended fails when nobody owns the self-led half.
Two failures cut across all four. The first is training people on seats that are not yet assigned, or on a tenant that has not been checked, so the first thing they learn is that Copilot cannot find their files. The second is choosing by cost per head when the hours people spend are the larger cost. This comparison also rests on one pilot and on our judgement about planning ranges, so use the numbers in the table as a starting point for your own quotes, not as benchmarks.
If your team splits into hesitant, confident and somewhere in between, talk through the split on a call and ivee will say which of the four formats fits, including when the answer is Microsoft's free material.
