AI assistant woven into Word, Excel, PowerPoint, Outlook, and Teams that drafts, summarizes, and analyzes across your work.
Last updated
Microsoft 365 Copilot is a seat-based enterprise purchase, and enterprise purchases fail for reasons that have very little to do with how good the model is. Almost everything that determines whether this deployment is judged a success happens before anyone types a prompt: who gets a licence, how your files are permissioned, what you promised the finance committee it would do, and whether anyone can tell afterwards.
So this page is about the buying decision. It assumes you already know it writes documents.
Clear this up first, because it is the most common source of confusion in this purchase. Microsoft Copilot is the free consumer assistant in Windows, Edge and the browser — chat, web answers, image generation, no connection to anything your company owns.
Microsoft 365 Copilot is a licence an administrator assigns to a named person, on top of a qualifying Microsoft 365 subscription, that puts the assistant inside Word, Excel, PowerPoint, Outlook and Teams and connects it to your organisation's own content. The grounding in your own files is the entire product. Everything else about it is available for free somewhere else.
There is a chat experience available to eligible Microsoft 365 users at no additional licence cost, and it is the single biggest source of disappointed expectations here. It is a chat window with enterprise data protection on it. It is not the in-app assistant. It does not summarise the meeting you missed, it does not answer questions from your mailbox, and it does not analyse the spreadsheet you have open.
The practical consequence is that a pilot run on the no-cost chat tier tells you nothing about whether the licensed product is worth buying, because the capability you are evaluating is absent from what you tested. Pilot the licensed version on a real team or do not pilot at all.
This is the part that surprises organisations and it deserves to be understood before rollout rather than during it. Copilot answers from content the user already has access to. That sounds like a safety guarantee, and it is — but it is a guarantee about permissions, not about intent.
Most large tenants have years of accumulated oversharing: sites shared with "everyone in the organisation" for convenience, files in a team space that half the company inherited access to, links generated for one recipient that became permanent. Nobody noticed, because finding those files required knowing they existed. A capable assistant over the same permissions is very good at finding things nobody knew they could reach. It has not leaked anything; it has surfaced an access model that was already wrong.
Treat a permissions and oversharing review as a prerequisite of the deployment rather than a follow-up item. Organisations that skip it tend to discover the problem through an incident, and the incident gets attributed to the AI.
Buyers evaluate this as one product and then use it as five, which is where the disappointment clusters. The pattern is consistent enough to plan around: the assistant does best where the input is a large amount of prose that needs compressing, and worst where correctness is binary and the source of truth is structured.
Meeting recaps and long mail threads are the strongest cases; the work is summarisation and the reader can verify quickly. Drafting from an existing document is solid. Slide generation produces a serviceable starting structure and a deck nobody would present unedited. Spreadsheet work is the weakest link and the one most often demoed, because the failure mode is a confident formula that is subtly wrong, which is exactly the error type a non-expert reviewer cannot catch.
Set expectations by app when you communicate the rollout. A user whose first three attempts were in Excel will conclude the product does not work, and they will not be entirely wrong about their case.
The distribution of value across an organisation is not uniform, and per-seat pricing punishes pretending otherwise. The people who gain most are the ones drowning in communication volume: managers in many meetings, roles that live in a mailbox, anyone who writes documents for a living. The people who gain least are those whose work is in a line-of-business application that Copilot does not touch, and they are frequently a majority of headcount.
Deploy to the roles where the argument is obvious, instrument what happens, and expand on evidence. Watch for the pattern where licences are assigned broadly, enthusiasm carries the first month, and usage concentrates in a small group by the third — that is the shape of a renewal conversation that goes badly, and it is entirely avoidable by starting narrow.
If your organisation does not actually live in Office — if the documents are in another suite, the chat is elsewhere, the knowledge is in a wiki — the grounding advantage evaporates and you are paying a premium for an assistant that cannot see your work. Notion AI makes the same argument for organisations whose content lives in Notion, and the comparison between them is really a question about where your documents already are.
If the need is individual rather than organisational — better writing, better reasoning, analysis of files a person uploads themselves — a frontier assistant like ChatGPT or Claude typically delivers more capability per pound, with the trade-off that it knows nothing about your tenant.
And if the goal is compliant coverage rather than capability, note that the no-cost chat tier already gives eligible users a protected place to work, which for some populations is the entire requirement. Buying licences for people whose real need was "somewhere sanctioned to paste things" is an expensive way to solve a cheap problem.
The strongest case by a distance: attendance is optional, decisions and actions are extracted, and the person who missed the call can verify the summary in seconds. This is also the capability that most reliably justifies a licence for a manager.
Summarising long threads, surfacing what needs a response and drafting replies in context. Value scales with communication volume, which is why the licence pays for itself in some roles and sits unused in others.
Producing a first version of a report, proposal or summary grounded in existing files rather than from a blank page. Reliable enough to change how the work starts, and still a draft that a subject-matter expert has to own.
Generating a deck outline from source material, which saves the tedious part and produces something nobody should present unedited. Useful as scaffolding; misleading if demoed as a finished artefact.
Locating the current version of a policy, a past proposal or a specification across the tenant. This is where an internal assistant beats any external one, and also where an untidy permissions model becomes visible.
Moving the pasting that already happens in personal accounts onto infrastructure with enterprise data protection. Worth separating from the rest of the business case, because the no-cost chat tier may already cover this population.
Microsoft 365 Copilot has a free-and-paid structure that is easy to misread. Copilot Chat is included at no additional licence cost for eligible Microsoft 365 users, but it does not connect to your Office apps — no reading mail, no meeting recaps, no analysis of your spreadsheets. The full in-app experience is a paid per-user add-on, quoted separately for business and enterprise customers, that sits on top of a qualifying Microsoft 365 subscription. Two things follow. The real cost is the sum of the underlying subscription and the add-on, so the add-on figure alone understates it for anyone not already licensed. And Microsoft has repeatedly revised both the per-seat rates and which plans qualify, so take current figures from Microsoft's own pricing page rather than from any summary, including this one.
Partly, and the free part is not the part people mean. Copilot Chat is included for eligible Microsoft 365 users, but it does not connect to your Office apps: it cannot read your mail, recap your meetings or analyse your spreadsheets. The in-app assistant that does those things is a paid per-user add-on on top of your existing subscription.
Not meaningfully. The capability you would be deciding on — grounding in your own documents, mail and meetings — is exactly what the no-cost tier lacks, so a pilot there measures a different product. Buy a small number of licences for a team whose work is communication-heavy and evaluate that instead.
It answers from content the user already has permission to open, so it does not bypass access controls. The practical risk is different: most large tenants have years of accidental oversharing that stayed harmless only because the files were hard to find. A capable assistant over the same permissions makes them easy to find. Run a permissions and oversharing review before rollout, not after.
It is strongest where the job is compressing a lot of prose — meeting recaps, long mail threads, drafting from documents you already have. It is weakest in spreadsheets, where a confident but subtly wrong formula is the characteristic failure and the reviewer least able to catch it is the one most likely to have asked. Set expectations per app or users will generalise from their worst experience.
Start with the roles whose day is communication volume: managers in many meetings, mailbox-centric functions, people who write documents for a living. Staff whose work happens in a line-of-business application outside the Office suite gain very little, and they are often the majority. Assigning licences broadly and hoping for uniform adoption is the pattern that produces a bad renewal conversation.
Decide before you buy, because retrofitting a measure to a renewal is not persuasive. Licence-level usage data will tell you who actually kept using it after the novelty passed, which is the most honest single signal available. Pair it with something concrete you expected to change — a meeting removed, a report no longer written by hand — rather than a self-reported time saving, which surveys inflate reliably.
They answer different questions. This is worth its premium only if the grounding in your own tenant is the point; if your documents, chat and knowledge live outside the Microsoft suite, you are paying for an advantage you cannot use. For individual capability — reasoning, writing, analysing files a person uploads — a frontier assistant generally gives more per seat, and plenty of organisations end up running both for different populations.
Full review coming soon.