Ask a business owner whether AI is helping, and the answer usually comes fast. Yes, obviously. Everyone's using it now, somewhere, for something. Ask them how much it's actually worth this quarter, in pounds, and the pause gets longer.

New research from McKinsey puts a number on that pause. Its 2026 "State of AI" survey of 1,719 professionals found that 88% of organisations now use AI in at least one part of the business, up from 78% a year ago. That's the adoption story everyone already knows. The less comfortable finding sits just beneath it: only 37% of those organisations can attribute any actual profit impact to their AI use, a figure that hasn't moved since last year's survey. Just 6% qualify as what McKinsey calls "AI high performers", businesses attributing at least 5% of EBIT to AI, with a significant measurable effect. Adoption climbed eleven points in a year. Return on that adoption stayed exactly where it was.

There's a second, quieter gap sitting inside the same data. 80% of individual AI users say the tools have made them personally more productive. That's a big number, and it's probably an honest one. Most people who use AI regularly do feel faster at parts of their job. The puzzle is why that feeling so rarely shows up as an organisational result. The likely answer is unglamorous: time saved on an individual task only becomes business value if someone deliberately decides what to do with it, redirect it, measure it, build a process around it, rather than letting it quietly disappear into the next task on the list. Most businesses haven't made that decision. They've handed out the tool and assumed the value would follow on its own.

The 6% who count as high performers appear to have done something different, not necessarily more AI, but a clearer sense from the outset of what "working" was supposed to look like, and a way of checking whether it actually did. That's a measurement discipline, not a technology advantage. It's available to a twenty-person business exactly as much as it is to a multinational, and arguably more easily, since there's less noise to see through.

There's a useful reassurance buried in the same report, too. 39% of respondents now expect their employer to cut jobs because of AI, up from 32% a year ago. It's the kind of stat that gets picked up and repeated as evidence something is coming. But McKinsey's own data shows that last year's actual AI-driven job cuts came in well below what that earlier group of respondents predicted. The fear is climbing faster than the reality underneath it. Worth knowing before it shapes a decision it shouldn't.

None of this is an argument against using AI, and it isn't a case for using more of it either. It's a case for treating "is this working" as a question worth answering on purpose, rather than one that gets inferred from a general sense that the business feels a bit faster than it used to. That's precisely what the Value dimension of a proper AI Accelerator Diagnostic is built to surface: not whether AI is present in the business, most now are past that point, but whether anyone has defined what a good outcome looks like and set up a way to know if it's arriving.

The businesses that end up in that 6% aren't the ones who adopted earliest or hardest. They're the ones who decided, before they started counting, what they were actually counting for.

Go wisely.

McKinsey, "The State of AI in 2026," via The Register, 25 August 2026