For most of the past two years, the story around AI adoption in UK businesses has been one of steadily rising confidence: more tools in use, more optimism about where they were heading. New data from Expleo's AI Pulse survey, reported this week by Industry 4.0 Today and Automation Magazine, breaks that pattern. The share of UK business leaders who believe AI is being used successfully inside their organisation has fallen from 80% to 67% over the past year. The number who see AI as primarily beneficial has dropped too, from 73% to 63%. Concern about cybersecurity now sits at 58%, and worry about the wider organisational disruption AI brings has climbed from 36% to 41%.

Read quickly, that looks like backlash - leaders souring on AI after an initial rush of enthusiasm. Read more carefully, it looks like something more useful: leaders asking AI to show its working.

There's a predictable shape to how organisations adopt any powerful new tool. First comes novelty - AI gets tried because it's available and the early wins feel remarkable. Confidence is high in this phase almost by default, because nobody has asked the harder question yet: what is this actually worth, and what is it actually costing us in risk? Expleo's numbers suggest a lot of UK businesses have now reached that harder question. The dip in confidence isn't evidence AI got worse. It's evidence the scrutiny finally caught up with the enthusiasm.

This is exactly the point in adoption where a proper diagnostic earns its keep, rather than another round of tool shopping.

Go Wisely's own AI readiness diagnostic looks across six dimensions rather than one - not just which tools a business has bought, but how ready its governance, its data, its people and its processes actually are to use them well. When we ran the full diagnostic with a 50-person global events and film agency, the value wasn't in confirming what they already believed about their AI use. It was in surfacing the gaps that confidence alone hides: where a tool had been adopted by one enthusiastic team member and nobody else, where cybersecurity exposure hadn't been thought through, where "we use AI" and "AI is actually changing how we work" turned out to be two different claims. That diagnostic fed into a six-session bespoke AI Accelerator training programme, which closed across 131 survey responses at 4.6 out of 5 satisfaction, with not one dissatisfied response.

The lesson isn't that this agency got lucky with a good result. It's that asking the evidence-based questions, before the confidence gap shows up in a board meeting, is cheaper and calmer than discovering it afterwards.

For any agency reading Expleo's numbers and recognising their own boardroom conversation in them, the useful response isn't defensiveness and it isn't panic. It's the same move Nathan makes with every new client: stop asking "are we using AI" and start asking "where, specifically, and how do we know it's working." That's an honest question, and honest questions are uncomfortable exactly once - the meeting where nobody has a clean answer. After that, they're just how a well-run adoption process works.

Cybersecurity and organisational disruption, the two concerns rising fastest in Expleo's data, aren't reasons to slow down out of nerves. They're reasons to slow down on purpose - the kind of pacing a proper diagnostic builds in from the start, rather than the kind that gets forced on you later by an incident or a board question nobody was ready for.

If this year's AI story was permission to experiment, next year's looks like it will be about the evidence to back up what you tried. Businesses that build that evidence deliberately, through a genuine readiness assessment rather than another dashboard of usage stats, will be the ones with a straightforward answer when the board finally asks the blunt version of the question Expleo just asked the whole country.

Go wisely.