The honest answer depends on two things nobody measures before buying: how often the task happens, and how expensive it is to check the output.
Microsoft 365 Copilot is a per-user, per-month addition on top of your existing licensing, and for most organisations it is the largest per-seat increase they have been asked to approve in years. That makes it a genuine decision rather than a default.
The question is answerable, but not by asking whether the technology is good. It is good. The question is whether it is good at tasks your specific people do often enough to repay the fee.
The two variables that decide it
Frequency is obvious. Verification cost is the one that gets missed, and it is the reason pilots disappoint.
If a tool drafts something in two minutes that would have taken twenty, but checking it carefully takes fifteen, the saving is three minutes rather than eighteen. For low-stakes work where errors are obvious at a glance, verification is nearly free and the saving is real. For work where an error is subtle and consequential, verification costs as much as the drafting, and sometimes more, because reviewing somebody else's plausible draft is harder than writing your own.
This is why the same tool genuinely transforms one person's week and does nothing measurable for their colleague.
Where the value reliably shows up
In pilots we have seen, the consistent wins are unglamorous:
Meeting summaries and action items. Frequent, low stakes, and errors are visible to everyone who was in the room. This alone is frequently the largest single contributor to measured value.
Finding things. Semantic search across documents, mail and chat, answering questions like "what did we agree with that client about the renewal". Saves minutes many times a day, and the time saved is almost invisible, which is why it rarely appears in business cases.
First drafts of routine correspondence. Standard replies, meeting follow-ups, straightforward internal communication.
Summarising long documents. A contract, a report, a long thread. Useful, with the caveat that it should orient you rather than replace reading the parts that matter.
And where it reliably does not:
Anything requiring current, authoritative accuracy that the reader cannot easily check. A plausible wrong answer is worse than no answer, because it is acted upon.
Specialist work with high consequences. The verification burden dominates.
Anything needing information the model cannot see, which includes most line-of-business systems unless you have done integration work.
The permission problem, which is the real prerequisite
Copilot answers from content the user already has permission to open. That sounds like a reassuring security property and it is, but it has an uncomfortable implication: it will surface things people technically can access and practically never would have found.
Most organisations have over-permissioned content. A SharePoint site shared with everyone years ago. A folder of salary data in a location where inheritance was broken. Files that are nominally accessible and effectively buried.
Copilot removes the burial. If somebody can open it, they can now ask a question that returns it.
This is not a reason not to deploy. It is a reason to audit permissions first, and the audit is work: find over-shared sites, fix broken inheritance, check anonymous links, and look specifically at where HR, finance and legal material lives. Budget for it. Organisations that skip this step have an uncomfortable fortnight.
What else is not in the licence fee
Training deserves a specific mention, because it is the variable most strongly associated with whether a pilot succeeds. The difference between someone who uses it for three things and someone who has integrated it into their day is almost entirely knowing what to ask. An hour of practical, role-specific training, with examples from their actual work, changes adoption substantially. Generic vendor training does not.
Running a pilot that tells you something
Decide the measure before you start. Most pilots fail to produce a decision because nobody agreed in advance what success would look like, so the result is enthusiasm from some users and silence from others.
Three measures that work:
Weekly active use per licence, which is brutal and informative. A licence used twice in a month is a licence to reclaim.
Self-reported time saved, collected the same way each week. Imprecise, and directionally useful across a group.
One specific task, timed. Pick something concrete, like producing the weekly client update, and measure it before and after on the same people.
Pick twenty to thirty users across three or four roles, including at least one sceptic. Run sixty days, not thirty, because the first fortnight is novelty.
The licence question
Nobody has to buy it for everyone, and most organisations should not start that way.
The roles where value shows up first are those doing high volumes of routine written work, people in many meetings, and anyone whose job involves finding information across a large document estate. The roles where it shows up last are deep specialists and people who work primarily in line-of-business systems rather than in Office.
Buy for the first group, measure, and extend based on what you measured rather than on who asks.
The short answer
For organisations already in Microsoft 365 with people who spend their day in documents, mail and meetings, Copilot is a defensible purchase for a subset of roles, provided you fix permissions first and spend something on training.
For organisations where most people work in a line-of-business system, or where the written work is specialist enough that every output needs careful checking, the honest answer today is to wait, run a small pilot, and measure rather than assume.
