There’s a pattern emerging across organisations — and it’s particularly visible in local government. London councils are rolling out Microsoft Copilot licences and presenting it as an AI strategy. Ealing trialled 250 seats. Enfield expanded to 635. Hammersmith & Fulham enabled Copilot for Edge so staff could “get familiarised”. Licences are purchased, pilots are launched, usage numbers are reported — and it begins to look like progress.
It isn’t.
The Tooling Trap
Rolling out Copilot is not the same as having an AI strategy. It is a tooling decision — not a set of business choices.
The numbers tell the story. Microsoft 365 Copilot has 15 million paid seats — but against 450 million commercial subscribers, that’s 3.3% penetration. A survey of 150,000 US respondents found that when employees had access to Copilot, ChatGPT, and Gemini simultaneously, only 8% chose Copilot. When it was the only option, 68% used it.
That 60-point swing tells you usage is driven by the absence of alternatives, not by preference. And preference is what predicts whether a tool actually changes how people work.
Forrester described enterprise adoption as “measured, even cautious,” with CIOs demanding outcome-led evidence upfront rather than generic productivity claims. This isn’t a Copilot problem. It’s a strategy problem — and it applies to every AI tool on the market.
Access Is Not Transformation
Many organisations mistake access for transformation. Licences are purchased, dashboards are shared, and it begins to look like something is happening. But a strategy answers harder questions: What is AI actually for here? Which workflows should it improve? What risks are acceptable? How will value be measured — not in seats deployed, but in outcomes delivered?
Without those answers, Copilot becomes another layer of activity: visible, modern, easy to talk about, but only loosely connected to results.
And here’s the part that rarely makes the business case: a Workday study found that 37% of time saved through AI is offset by rework — correcting, verifying, and rewriting outputs. Every ROI projection built on gross time savings ignores this tax. Used well, these tools reduce low-value admin and make knowledge work easier. Used badly, they sit on top of poor processes and vague expectations.
Start With The Workflow
The correct first question is not “which AI vendor should we deploy?” It’s “which specific process costs the most per unit of output, and what would it mean to improve it by 40%?” The answer selects the use case. The platform decision follows.
The organisations that benefit most won’t be the ones that switched it on first. They’ll be the ones that knew what they wanted it to change.
If you can’t answer “what, specifically, did this change?” — you don’t have a strategy. You have a subscription.