Nearly three years after Microsoft 365 Copilot became generally available for enterprise customers, opinion on it inside organisations tends to split along a predictable line. Some teams describe it as part of the furniture, used every day for meeting summaries, first drafts and digging through long email threads. Others bought licences, watched usage fade after the first month and are now wondering whether to renew. Often both views exist inside the same company.
The difference usually has less to do with the product than with how it arrived. Treated as a licence purchase, Copilot tends to disappoint. Treated as a change to how information is stored, secured and used, it has a much better chance.
The case for the sceptics
It’s worth taking the disappointed camp seriously, because their complaints are often fair. Copilot’s output is only as good as what it can find, and in many Microsoft 365 tenants what it can find is a mess: duplicate versions of the same document, abandoned Teams sites, and SharePoint libraries where the current policy sits next to three superseded drafts with similar names. Ask Copilot a question in that environment and it will answer confidently from whichever file it reaches first.
Then there’s the cost question. Copilot is priced per user per month on top of existing Microsoft 365 licences, and for many staff the time saved is real but hard to measure. A finance team can see the invoice clearly. The benefit shows up as slightly quicker emails and fewer minutes spent searching, which rarely appears on anyone’s dashboard. When budgets tighten, that asymmetry doesn’t help.
Some sceptics also point out that the most useful features have spread beyond the paid product. Microsoft now offers Copilot Chat to business users without a full Copilot licence, and many people find that web-grounded chat covers a good share of what they wanted. That’s a reasonable argument for a smaller, targeted deployment rather than licensing everyone.
What the successful organisations did first
The organisations getting clear value from Copilot tend to share a less glamorous starting point. Before switching it on widely, they looked hard at their data and their permissions.
This matters because of how Copilot works. It respects the permissions a user already has, which sounds reassuring until you remember how loose those permissions often are. Files shared with “everyone in the organisation” years ago, sites with inherited access nobody reviewed and sensitive folders that were only ever protected by being hard to find all become much easier to surface once an assistant is doing the searching for you. Oversharing was always there. Copilot just removes the obscurity that was quietly keeping it in check, which is why organisations with large estates or stretched IT teams often bring in advisers who prepare data and permissions before rollout rather than finding the problems live.
The fix is mostly housekeeping, done properly. That means reviewing broad sharing links, tidying site membership, archiving stale content, and applying sensitivity labels through Microsoft Purview so that confidential material is marked and handled accordingly. Microsoft has added tooling to help, including SharePoint features for finding overshared sites and restricting what Copilot can search while the clean-up happens. The organisations that skipped this step are the ones most likely to have paused a deployment after someone found a salary spreadsheet in a Copilot answer.
Data quality is the other half. Copilot works noticeably better when there’s a clear, current version of important documents and when teams store work in shared locations rather than personal drives and email attachments. That’s an information management habit more than a technical setting, and it takes time to build.
Adoption is the part people underestimate
Even with clean data, usage doesn’t grow on its own. People who’ve never been shown what Copilot is good at tend to try it once on something it’s bad at, conclude it’s overhyped and go back to their old habits.
The more successful rollouts treat adoption as a programme in its own right. They start with specific roles and specific tasks, such as account managers preparing for client meetings or project leads summarising long Teams channels, rather than handing licences to everyone and hoping. They find people in each team who are keen, give them time to experiment and let them show colleagues what works. They also measure something beyond active users, such as time spent on a particular recurring task before and after.
None of this is unique to Copilot. It’s the same change management that any new tool needs, and many organisations skipped it because Copilot arrived inside software people already used every day. That familiarity turned out to be misleading.
The harder question now is what happens as Copilot shifts further towards agents that carry out tasks on a user’s behalf. A chat assistant that surfaces the wrong document is embarrassing. An agent that acts on it, by sending it to a client or updating a record, raises the stakes considerably, and the permissions clean-up that many organisations postponed will stop being optional.