The pilot started with 25 licenses and a welcome email. By week two, half the users had tried Copilot a couple of times. By week six, the report still showed "active users," but nobody could name a single task that was getting done faster. When renewal came up, the finance manager asked a simple question: what exactly are we paying for? Nobody had an answer.
Copilot adoption in business rarely falls apart because of the technology. It falls apart earlier, in decisions that look minor, like handing out licenses without a use case or switching the tool on over permissions nobody has reviewed in years. It's a story that plays out often, and it can almost always be avoided. Below are the four mistakes that show up most often in Copilot pilots, what the data says, and what to do differently so yours ends with a clear decision.
What the data shows about Copilot pilots
In Gartner's 2025 Microsoft 365 and Copilot survey of 187 IT leaders, 40% said they were running a pilot and another 19% were planning one. Among organizations that had finished a pilot, only 5% were moving to a larger deployment that year, according to what analyst Dan Wilson presented in Sydney. Nearly half of those who had tried Copilot rated it "some value, shows promise."
That 5% doesn't mean 95% failed. Plenty of companies kept going with small groups, waiting to see where it paid off. What it does make clear is that finishing a pilot isn't enough to justify the next step.
Back in 2024, Gartner had also noted that the value Copilot delivers tracks closely with information management maturity, and that user enablement and security mitigation take more effort than expected. You'll find both behind most pilots that stall.
Mistake 1: licenses without use cases
The most common pattern is giving licenses to whoever raised their hand first. That measures enthusiasm, not value. A month later, the enthusiasts are still using it to draft emails and everyone else has opened it twice.
A useful pilot works the other way around. First, pick two or three concrete tasks per department, such as preparing the monthly sales report, summarizing follow-up meetings, or analyzing variances in a cost spreadsheet. Then pick the people who do those tasks every week. If you can't name the task, you don't have a pilot yet.
Microsoft's scenario library has examples by industry and role that are good conversation starters. You still have to map them to your own processes. A generic "marketing" scenario means nothing to a sales team selling industrial maintenance services.
Mistake 2: training once (or never)
Many companies assume people will pick up Copilot on their own because it lives inside Word and Outlook. They won't. Writing good instructions, knowing which sources Copilot draws on, and spotting when an answer relies on old or wrong data are skills people have to practice.
The opposite approach fails too: a two-hour session on day one and nothing after. Weeks two and three are the critical ones, because that's when people hit a mediocre result and slide back to their old way of working if nobody helps. Wilson put it this way: Copilot changes so quickly that fixed step-by-step guidance stops working, and evaluation has to be continuous.
It's also worth knowing that Microsoft's FastTrack program for Copilot offers remote technical guidance but lists end-user training and change management planning as out of scope. That part falls on the company, or on whoever is helping it.
What tends to work:
- A first session using the team's real documents, not catalog examples.
- A Teams channel for questions, with someone who answers the same day.
- One or two go-to people per department who share what worked for them.
- A second session three or four weeks in, to fix bad habits and show cases that already worked inside the company.
For executives, who usually decide whether the pilot continues, we wrote about how we run a 4-hour executive Copilot workshop. Once leadership has used the tool on their own work, the budget conversation gets much more concrete.
Mistake 3: messy permissions
Copilot answers with whatever information each user can already access. If a SharePoint site with payroll data was opened to the whole organization three years ago, Copilot doesn't create the problem, but it surfaces it in seconds. One case like that is enough to freeze the pilot while someone scrambles to fix permissions.
Gartner describes the same thing: deployment plans tend to assume clean data, which is rarely the case, and the risk mitigation effort ends up much larger than planned.
Microsoft recommends completing the Copilot Optimization Assessment before deploying, to review data governance and security maturity. On top of that, SharePoint Advanced Management is included with the Copilot license and has tools to find and fix oversharing. You don't need to clean up all of SharePoint before you start. You do need to review sensitive libraries and sites open to everyone before assigning the first license. We cover that stage in our guide to Microsoft 365 Copilot implementation in 90 days.
Mistake 4: measuring usage and calling it results
The admin center report tells you how many people used Copilot, not whether it helped them. Wilson pointed out that the default setting counts a single use in 28 days as adoption. He doesn't consider that adoption: he suggests at least one use per week as a minimum, and ideally one per day, to justify the license cost.
Before you start, measure how long the chosen tasks take today. Without that baseline, any number at the end of the pilot is an opinion. We explain how in AI productivity: how to measure it.
What a pilot that reaches a decision looks like
| Aspect | Pilot that stalls | Pilot that decides |
|---|---|---|
| Who gets a license | Volunteers and managers | People who do the chosen tasks |
| Use cases | "Just try it" | Two or three per department, in writing |
| Training | One kickoff talk | Hands-on session, follow-up and go-to people |
| Permissions | Reviewed when a problem shows up | Sensitive libraries reviewed beforehand |
| Measurement | Active users | Time per task against a baseline |
| Outcome | "People seem to like it" | Continue, extend to similar roles, or stop |
About that last row: stopping is a good outcome too, if it's well grounded. For some roles, Copilot Chat, at no additional cost for users with Microsoft 365, is plenty. Knowing that keeps you from paying for licenses out of habit.
Frequently asked questions
How many people should a Copilot pilot include?
It depends on company size, but for a mid-sized company 15 to 30 people from two or three departments is usually enough. The number matters less than everyone having a defined use case.
Does Microsoft provide Copilot training?
Microsoft publishes free adoption material, such as the Success Kit and introductory sessions. Its FastTrack program, however, excludes end-user training and change management strategy, so the company has to organize that part.
What should I do if my pilot has already stalled?
Don't buy more licenses yet. Pick two concrete tasks, measure how long they take today, run a hands-on session with the team that does them, and review their permissions. In four to six weeks you'll have data to decide.
Make your pilot end in a decision
If you're about to start a pilot, or already have one that isn't moving, Disrupsoft can help with an adoption assessment to find the highest-impact scenarios, activation and configuration of Copilot in your Microsoft 365 environment, hands-on training for your users, and ongoing support so usage keeps improving over time. Take a look at our Copilot and Microsoft 365 service and let's talk.
