There’s a moment, somewhere between month four and month seven of a Microsoft 365 Copilot rollout, when a CFO asks the IT director a fairly specific question. The question is some variant of: we paid how much for these licences again?
The honest answer is usually a number with three commas in it. The honest follow-up question, which the IT director sometimes asks back and sometimes doesn’t, is harder. Where, exactly, is the productivity we promised?
Eighteen months in, the Copilot story across the UK enterprise market has split into two distinct camps. The first camp has, broadly, no answer to the productivity question. Their users like Copilot in a vague, polite way. They use it to summarise meetings and draft emails. Adoption metrics look fine on paper. But nobody at executive level can point to a specific business outcome that has measurably moved. The licence renewal conversation is going to be difficult, and the IT director knows it.
The second camp is, frankly, having a different experience entirely. They can point to roles where work that used to take a day takes two hours. They have case studies with ROI numbers that look made up but aren’t. They are renewing without drama and expanding their licence base. Something is different about how they deployed.
That something is almost never the technical implementation. The technical implementation is mostly the same across both camps; Microsoft has done the work to make it boringly consistent. The difference is upstream of the technology, in a decision the second camp made early and the first camp didn’t.
The second camp picked roles where the floor was low, not where the ceiling was high.
This sounds obvious phrased that way. In practice, it is the opposite of what most organisations instinctively do. The instinct is to give Copilot to the senior people first. Senior people have the most expensive time. Senior people are the most visible. Senior people will report back. So the rollout goes to the executive team, the senior managers, the heads of department.
Senior people are also, almost without exception, already operating close to their ceiling. They know their domain. They write quickly. They’ve spent twenty years learning to summarise their own meetings. The marginal productivity gain from giving a senior person a fluent writing assistant is real but small. They use it for an hour a week and forget about it for the rest.
The roles where the floor is low are different. They are the people who write a lot of structured but unfamiliar text: service desk agents writing incident reports, junior consultants writing client emails, sales development reps writing follow-ups, anyone in a customer-facing role at the beginning of their career. These people are often producing work at the seventy-fifth percentile of what’s possible for that task, because they haven’t yet built the experience to do better. Giving them a tool that nudges them toward the ninetieth percentile is a measurable, repeatable, visible productivity gain. And, less talked about but more important, it lifts the floor of what the organisation produces. The worst version of a customer email goes from embarrassing to acceptable.
See also: How Technology Drives Business Innovation
The AI consulting practices working in the Microsoft stack that are doing this well now have a recognisable methodology. They don’t start with the executive team. They start by identifying three or four roles where the work is high-volume, structurally similar, and currently being done at varying levels of quality. They deploy there first. They measure. They build the business case from the bottom up. Only then do they roll out widely, by which point there’s an actual answer to the CFO’s question.
There’s a secondary effect worth mentioning. Organisations that did the executive-first rollout, and now can’t explain the productivity gap, are increasingly bringing in outside consultants to help them retrofit the use-case work they should have done in the first place. The market has noticed. There’s a perfectly respectable practice category emerging around what is essentially Copilot rescue work, finding the use cases, retrofitting the measurement, and rescuing the renewal conversation.
The point is not that Copilot doesn’t work. It works. The point is that nobody told most organisations the dirty bit, which is that getting value out of a generative AI tool inside a large business is not primarily a technology problem. It’s a question of where, in your organisation, the gap between what’s currently produced and what’s possible is widest. That gap is almost never at the top.
