What Copilot actually does
Microsoft 365 Copilot is a paid AI assistant layered on top of the Microsoft 365 apps you already use. Inside Word, Excel, PowerPoint, Outlook, Teams and SharePoint, it drafts and rewrites documents, summarises long threads and files, spots trends in a spreadsheet, and transcribes, translates and summarises Teams meetings[1]. It also handles some of the repetitive admin work nobody enjoys - generating a first-draft report, drafting a follow-up email, pulling together a summary of what was decided in a meeting you missed[1].
What it is not: a separate database, a search engine with its own index of your company, or something that "knows" more than the people using it. It works inside the apps you already have, on the files and messages those apps already show a given user.
Real examples from Microsoft's own case studies give a clearer sense of scale than the feature list does. One marketing team went from three hours to as little as 30 minutes drafting a campaign brief. A customer-feedback review that used to take three or four hours now takes under one. A content team cut production time from weeks to days[1]. None of those numbers will match your business exactly, but they're the shape of what changes: less time on the first draft, more time on the judgement call that comes after it. The reason that lands is where the week already goes: McKinsey's work on knowledge work put roughly a fifth of the working week into searching for internal information and chasing the colleagues who have it[2].
What it costs, for a real team
Copilot is not sold on its own any more. It sits on top of a Microsoft 365 Business plan, and for new customers it now only comes bundled with one[3]:
- Business Standard + Copilot: about $23.50 per user, per month, billed annually.
- Business Premium + Copilot: about $32 per user, per month, billed annually.
- Standalone Copilot Business add-on, for businesses that already hold a qualifying licence: $18 per user, per month, promotionally through 30 September 2026, reverting to $21 after[3].
For a 10-person team on Standard, that's roughly $235 a month, or about $2,820 a year, on top of whatever the base Microsoft 365 licence already costs. On Premium it's closer to $320 a month. Both bundles cap out at 300 users per tenant[3] - past that you're into Enterprise licensing, a different conversation.
That's the whole cost. There's no separate infrastructure to run, no server to patch, no consultant needed to "install" it. The only real cost beyond the licence is the one nobody puts in a pricing table: the time it takes your team to actually change how they work.
Why the return varies so much
Forrester's commissioned study of small and medium businesses using Copilot modelled three scenarios over three years: a low-impact case at 132% ROI, a medium case at 243%, and a high case at 353%, based on a survey of over 200 companies with up to 300 employees[4]. The same study reported new-hire onboarding accelerating by 25%[4].
Two things are worth being honest about before you take that range as gospel. First, it's commissioned by Microsoft and projected - built from financial modelling across risk scenarios, not measured against actual before-and-after deployments[4]. Treat it as a framework for what's plausible, not a guarantee for your business. Independent tracking outside the vendor ecosystem points the same way with narrower numbers: Stanford's AI Index finds business adoption climbing fast, and the productivity gains measured in controlled studies real but smaller than vendor projections[5][6]. Second, and more useful: the range is wide mostly because of adoption, not because of anything the licence does differently at different price points.
Microsoft's own 2026 Work Trend Index, based on a survey of 20,000 full-time knowledge workers across ten countries, found that when a manager actively uses AI themselves and is visibly seen doing it, employees report a 17-point lift in the value they get from it, a 22-point lift in critical thinking about how they use it, and a 30-point lift in trust in agentic AI tools[7]. Workers whose manager creates space to experiment are 1.4 times more likely to become high-frequency users[7].