"Everyone is using it. Nobody has agreed on what for."— Office Man
What is it?
Microsoft Copilot is AI built into Microsoft 365. It sits inside Teams, Outlook, Word, Excel, and PowerPoint — the tools you already use. It is powered by OpenAI's models but operated through Microsoft's environment, under your organisation's data agreements. It can summarise meetings, draft emails, recap documents, generate slide decks, and analyse spreadsheet data. It does not have access to everything in your organisation — only to what you have access to. Your files, your emails, your meetings. Not your colleagues' unless they have been shared with you.
Why now?
Microsoft began rolling Copilot out to enterprise customers in 2023 and has been expanding it since. It is expensive — the licence cost is significant — and organisations are trying to justify it. Some people find it immediately useful. Others have access and barely use it. The rollout has been accompanied by ambitious claims from Microsoft and more mixed experiences in practice. The result: a lot of noise about a tool that many people have but fewer people understand.
The human angle
The honest version of what Copilot does best is mundane. Meeting notes. Email drafts. Document summaries. These are not glamorous. They do not sound like a revolution. They are, however, tasks that consume a disproportionate amount of time for most knowledge workers. The meeting recap feature alone — knowing what was discussed and agreed, without having to write it up yourself — has real value for anyone attending many meetings. The gap between what Copilot is marketed as and what it is good at in practice is mostly a gap between ambition and useful routine.
What to do
Start with meeting summaries in Teams — this is where most people find consistent value. Try email drafts in Outlook for routine correspondence. Use document summaries in Word when you need to get across a long file quickly. Ask your IT team what Copilot has been given access to: not every deployment is the same. Leave the more complex automation features until the basics are producing results. The tool works best when you treat it as a first-draft machine and apply your own judgement to what it produces.