We run Microsoft 365 Copilot pilots for UK businesses every month. The ones that succeed look nothing like the ones that stall. The tech is identical. The difference is everything around the tech.
Here are the four reasons pilots fail, in order of frequency.
1. The data estate is a mess
Copilot retrieves from SharePoint, OneDrive, Teams and Exchange. If your SharePoint is an oversharing disaster, Copilot will helpfully surface salary spreadsheets to the whole company in response to innocent questions.
Before any pilot, do a SharePoint sharing audit. Default sharing for new sites to “specific people”. Run the SharePoint Oversharing Report in the Microsoft 365 admin centre. Tag anything genuinely confidential with sensitivity labels. Then, and only then, enrol users in Copilot.
This is the single biggest blocker we see. Not because anyone configured SharePoint wrong, but because it has grown organically for five years and nobody has ever audited it.
2. Licensing the wrong people first
Licence the C-suite and the project team. That is the typical first mistake.
The C-suite do not have the time or inclination to learn a new tool. The project team do not have the document volume to feel the benefit. Neither group will come back with adoption stories.
Licence the people who write long emails, summarise meetings, do first drafts of documents, and live in PowerPoint. That is usually your marketing team, your sales ops people, and your senior managers with lots of reports. Give them Copilot, train them for an hour, and check back in a fortnight. They will tell you how good it is, loudly, and the rest of the business will queue up.
3. No prompt library
The difference between “Copilot is rubbish” and “Copilot saves me five hours a week” is usually a single week of learning to prompt it properly.
Build a prompt library. 20 to 30 prompts, organised by role and task. “Summarise this meeting, action items only, with owners.” “Draft a project status email from this Teams chat.” “Turn these bullet points into a one-page brief.”
Share the library in a pinned Teams channel. Encourage staff to add their best prompts. The compound effect is significant.
4. No measurement
If nobody is measuring, nobody is adopting. Microsoft’s Copilot Dashboard tells you exactly who has used it, for what, and how often. Pull the report weekly during a pilot.
More important than the numbers is asking users, every fortnight, three questions. What did you use Copilot for? What did not work? What would make you use it more? The answers drive the next two weeks of training and prompt library additions.
The pattern that works
Two-week readiness phase: tenant audit, SharePoint hygiene, licence plan, target user list, training materials. Four-week pilot with 20 to 40 carefully chosen users. Weekly stand-ups. Prompt library grown in-pilot. Measured adoption, measured time-saved.
After six weeks, you have real evidence of what works, you have a trained cohort who are advocates, and you have a remediation list for anything that broke. Only then do you go wider.
What to not do
Do not buy a hundred Copilot licences, email the company, and hope. It does not work. Adoption will plateau at 15 percent in month three and the pilot will be declared a failure in month six.
Do not try to ban it while figuring out a strategy. Staff will paste customer data into public chatbots instead. Give them the sanctioned option, fast.
Our AI team runs Copilot readiness reviews on a fixed-fee. Two weeks, a proper plan, and a decision you can defend to the board.