The Gap Nobody in My Industry Wants to Look At

I want to give you one number and ask you to sit with it. 88% of companies are using AI. Around 6% are getting real value out of it.

That is McKinsey's State of AI 2025, nearly 2,000 organisations across 105 countries, surveyed this summer. It is the most useful number I have, because it kills the thing everyone hides behind. Usage. Once you separate usage from value, "we're using AI" stops meaning anything at all.

Most people in my industry would rather talk about the 88%. They can sell you something for the 88%. I spend my time in the gap between the two numbers, because that is where the actual work is, and that is where every business I walk into is stuck.

The value gap, in plain numbers

BCG studied more than 1,000 companies this year. 26% are getting real value from AI. 60% have seen little to no benefit, despite real money going out the door (BCG, The Widening AI Value Gap, 2025).

Read those two findings together and the conclusion is uncomfortable. The same investment produces a result for one company and nothing for the next. Whatever explains that difference, it is not the model. Everyone has access to the same models.

BCG splits the challenge three ways: 10% algorithm, 20% technology and data, 70% people and process. The whole industry crowds onto the 10%. New model launches, capability threads, the conference circuit. The 70% that decides whether any of it survives contact with a real business is the part nobody posts about, because there is no product to launch and no dashboard to screenshot. It is pulse surveys, resistance mapping, champion training, governance design. Slow, careful, human work.

That 70% is the difference between the 6% and everyone else.

What it looks like in the room

I sit in a lot of boardrooms, and the pattern barely changes.

An MD shows me the subscriptions the company has been paying for. Copilot, a sector tool, a few individual Claude plans. Four months in. Usage is low. Nobody is quite sure what any of it was meant to fix.

I ask one question. What problem were you solving when you bought this?

Usually silence. Sometimes three different answers from three people who have never compared notes.

That is the diagnosis problem, and it is upstream of every tool decision. Before anything gets bought, somebody has to have answered: what specific friction are we removing, for which people, and how will we know it worked.

Here is the exercise I run before a single pilot. Everyone who will touch the system writes down, by hand, exactly how they do the task today. Handwritten, not a template. The ones who cannot describe their own process in clear steps are not ready to hand it to a machine. That one exercise has stopped three planned pilots cold, because it drags the invisible thing into the light: a manual decision step, a process that changes by client, a workaround that has become a habit nobody names.

The tool was never the problem in those cases. The diagnosis was.

Why the resistance is rational

Employee resistance gets blamed for a lot of stalled AI. The usual fix offered is better communication, more enthusiasm from the top.

That misses what the resistance actually is. When you work somewhere that bought a tool without explaining what it is for, or what happens to your role once it works well, holding back is a sensible response to ambiguity. People are not resisting the technology. They are resisting being kept in the dark about what it means for them.

Which changes the fix. The answer is removing the ambiguity, not adding more cheerleading. An AI manifesto that says, in plain terms, how the time saved gets used. A clear line on how roles evolve. A champion in the team people can actually go and ask. Build that cultural layer before the tools land, and resistance drops on its own. Most companies build it afterwards, if they build it at all.

What the 6% actually do

McKinsey's high performers are not waiting and they are not lucky. They redesign their workflows around AI rather than layering it on top. They have leaders who use the tools themselves, not just sponsor them. They put real budget behind capability, not just licences.

The first one is the one that hurts. Redesigning a workflow means documenting how the work happens now, finding the friction, and placing AI at the exact point in the line where it pays off. Slow, unglamorous, and the reason their numbers look nothing like everyone else's.

Where this leaves you

The honest version of "AI adoption is stalling" is not that the technology fails. It is that the 70%, the people and process work, keeps getting skipped for the 10% that demos well.

One problem. One pilot. One measurable outcome. Fix the process first, then make it faster. Two quick wins to build belief with the people who were most sceptical, one strategic win to prove it scales, then run it again.

The wow moment matters here. The first time someone watches a task they have done by hand for three years happen in seconds, their resistance is gone and their imagination switches on. You cannot reach that moment without the groundwork. The groundwork is the 70%.

Skills that stay when we go.