Quick exercise. Count the AI licences your company is paying for right now. Then count the roles, processes or team structures that changed because of them.
For most operators I talk to, the first number is somewhere in the dozens. The second is a hesitant zero.
That’s the story of the last eighteen months in this industry, and honestly in most industries. We bought the tools. We ran the pilots. Someone did a lunch-and-learn. And the company on the other side of all that activity is structurally identical to the one that existed before.
I’m not guessing at this. MIT’s NANDA group looked at 300 public AI deployments last year and found that 95% of enterprise pilots delivered no measurable P&L impact, even though more than 80% of organisations had piloted tools like ChatGPT or Copilot. Their phrase for it was “high adoption, low transformation.”
McKinsey’s latest global survey is even more telling because it’s the second year in a row of the same result. Nearly nine in ten organisations use AI regularly in at least one function. Only 37% report any impact on enterprise EBIT at all, essentially unchanged from twelve months earlier, and the share of what McKinsey calls high performers is stuck at 6%.
Here’s the number in that report I can’t stop thinking about. 80% of respondents say AI has improved their own productivity. Eight in ten. And yet the company-level figure hasn’t moved. Everybody is faster and the business isn’t.
So nearly everyone has it. Almost nobody is getting much from it. And the interesting question is why, because it clearly isn’t the models. The 6% and the 94% are using the same ones.
Here’s what I think actually happened, based on watching it happen.
AI arrived inside operators the way most software arrives: as a purchase. Procurement, a licence, a rollout email. That framing quietly decided everything that followed. A purchase gets measured on usage. So the reporting was about seats activated and prompts per week, and by that measure the whole thing looked like a success. People were using it. They still are.
Meanwhile the actual work changed underneath. Tasks that used to take an afternoon started taking twenty minutes. Some tasks stopped needing a person at all. And because nobody had been asked to notice this, nobody did. The job descriptions stayed. The team sizes stayed. The headcount plan for next year got built on last year’s assumptions about how long things take.
McKinsey’s data is blunt on this. Nearly three-quarters of high performers say they have fundamentally redesigned workflows because of AI. Among everyone else, it’s one in four. Most companies bolted the tools onto existing processes and left it there, and the report’s own conclusion is that the limiting factor is no longer the technology but the organisation’s ability to absorb change.
Bolted on. That’s the phrase. That’s what eighteen months of buying looks like from the inside.
And it’s not one tool, it’s a whole stack of them, each one bolted on the same way. An MCP connection that lets an agent pull straight from the CRM instead of a support rep switching screens. A skill that encodes a team’s compliance checklist so it doesn’t have to be re-explained every time. An agent that runs game-testing checks with the persistence of a QA team that never gets tired. A CEO’s AI assistant pulling together board notes and contracts before a call. None of that is hypothetical anymore, it’s live in operators today. What’s rare is any of it showing up on a scorecard, a role description, or a headcount plan.
There’s a second finding I find almost funny. A year ago, 32% of companies expected AI to shrink their headcount. When McKinsey went back and asked what actually happened, only 14% saw a decline. Two-thirds saw no change at all. Everyone predicted the org chart would move. Then nobody moved it.
And the MIT report found something I think matters even more for us. Employees in over 90% of firms were using personal AI tools regardless of whether the official pilot was working. (Forbes, August 2025) So the transformation is happening. It’s just happening one person at a time, invisibly, with no one owning the outcome, no one checking the output, and no one adjusting the structure around it.
In a regulated industry that should make you slightly nervous. A compliance summary drafted by a tool nobody signed off on. A promo text that went out because it looked fine. Not a disaster, usually. Usually.
What worries me isn’t the tools. I like the tools. What worries me is that we’ve built a layer of AI(?) into the day-to-day work of the business and left the organisation chart from 2023 sitting on top of it.
I could be off on how fast this bites. I don’t think I’m off on the direction. By next summer most of us will have roughly the same AI capability. The difference will be who redrew the chart to match.
That’s the part I’ll be taking on at the Acceler8 Lab at SBC Summit Lisbon: what AI could do now, how it rewrites how will we work in the next summer and the ones coming after it.
September 30, 15:00, stand C416. Register now: https://gr8.tech/acceler8-lab/
Bartek Borkowski, Founder of PlayPatrol