Microsoft Copilot

Microsoft Copilot is the most widely licensed AI assistant in enterprise and one of the least loved, and both facts have the same cause: it is sold to organisations rather than chosen by users.

The case for it

If your organisation runs on Microsoft 365, Copilot has access to something no competitor can get: your actual work. Your mail, your files, your meetings, your chats, and — critically — the permissions model that governs who can see what.

That last part is the genuinely hard engineering. An assistant that can answer “what did we decide about the Henderson contract” by reading across Teams, Outlook and SharePoint, while respecting the same access controls the underlying systems enforce, is a different product from one you paste documents into. Microsoft Graph is the moat, and it is a real one.

The meeting features are where users most often report actual value. Recap, transcript, action items and the ability to ask questions about a call you missed are straightforwardly useful and require no behaviour change.

Where it falls down

Output quality is inconsistent in a way that is hard to predict and therefore hard to trust. The same question phrased two ways produces materially different answers; a summary is excellent one day and superficial the next. Users learn from a few bad results to stop trying, which is how a licensed seat becomes an unused one.

The retrieval layer is the usual culprit rather than the model. Copilot has to decide what of your organisational data is relevant before generating, and when that step goes wrong the answer is confidently built on the wrong documents. There is no signal distinguishing a well-grounded answer from a poorly grounded one.

The surface fragmentation is a real usability problem. Copilot in Word, Copilot in Teams, Copilot in Windows, Copilot in Edge and the standalone consumer Copilot are different capabilities under one name, and users reasonably form an impression of the product from whichever one they touched first.

Adoption is the number that should worry buyers. Licensed seats routinely outnumber weekly active users by a wide margin, and the gap is usually not a training problem — it is that the tool did not repay the effort the first three times.

Data governance cuts both ways

The permissions-aware retrieval that makes Copilot useful also makes it an excellent surfacing mechanism for permissions you did not know were wrong. Organisations with years of over-shared SharePoint sites discover this quickly: Copilot will cheerfully summarise a document the user technically had access to and was never meant to find.

This is not a Copilot defect, it is an existing problem made visible. But it means a serious deployment starts with a permissions audit, not with licence assignment. Budget for that work.

Pricing

A free consumer tier exists and is a different product from the enterprise one. Business Copilot is a per-seat Microsoft 365 add-on at a price that is significant at organisational scale.

Given the adoption gap, the sensible procurement approach is a genuine pilot with a defined measurement window — not a department-wide rollout on the assumption that usage follows licensing. Measure weekly active use, not enthusiasm in the kickoff meeting.

Who should use something else

Individuals and small teams almost always get more from ChatGPT or Claude at lower cost, because the Graph integration that justifies Copilot’s price is an organisational asset rather than a personal one.

Organisations on Google Workspace have the mirror-image decision with Gemini, and the same caveats apply.

For the free options worth trying first, see best free AI tools.