AI Adoption Tripled. Depth Didn’t.

AI adoption in UK business has almost tripled in under three years. On paper that looks like a transformation. The Office for National Statistics, which released the figures on 20 July, reaches for a blunter word: shallow.

The headline holds up. Among firms with 10 or more employees, use of at least one AI technology rose from around 12% in late 2023 to around 35% by June 2026. Beneath that curve, though, almost nothing deepened. The distance between using AI and running on it is the part your board rarely hears about.

The AI adoption number that flatters the dashboard

Breadth is the easy figure to report, and it flatters whoever reports it. A third of firms now touch AI in some form. Larger firms lead, at 49% of businesses with 250 or more staff against 28% of the smallest.

Counted this way, though, AI adoption measures reach, not capability. The ONS is unusually honest about the limit of its own headline. Its measure treats every reported use as equivalent, and cannot separate light, user-level dabbling from embedded, production-level work. Reach climbed. Whether the work itself changed is a different question, and the release answers it directly.

What the depth numbers actually say

Start with how many distinct AI technologies a typical adopter runs. Since late 2023 that figure moved from around 1.4 to around 1.6. Nearly three years of noise, and the average adopting firm added a fraction of a single tool.

The intensity data is starker still. Of firms using any AI, only 10% call that use extensive. Just 15% say more than half their people use AI in daily work. Large language models top the mix at 18% of firms, with visual content tools next at 16%. Most organisations run one or two narrow use cases at the edge of the business, nowhere near its core.

The shape of AI adoption is easy to see at a glance:

• reach: from around 12% to around 35% of firms since late 2023
• technologies per adopter: from around 1.4 to around 1.6
• firms using AI extensively: 10%
• firms where over half of staff use AI daily: 15%

One line rises steeply. The other three barely move.

AI adoption is not integration

Here the distinction earns its keep. A tool in the building is not a process rebuilt around it. The ONS finds that the most common purpose of AI, reported by close to 60% of firms, is improving existing operations rather than creating new products or opening new markets. Firms bolt AI onto the way work already happens. They rarely change what the work is.

The workforce data agrees. Most firms report no change to headcount from AI at all, a pattern the Bank of England’s Decision Maker Panel also found in January. Efficiency is the ambition. Structural change is not yet the result. None of that spells failure. It is a signpost, and it says AI adoption as counted is a starting line dressed as a finish.

Why the gap is a capability problem

If tools are cheap and often free, why does depth stall? The release points at people, not technology. When firms name a barrier, they name a lack of expertise most, rising to around 18% among those with 100 to 249 staff. Their most common answer is training or retraining existing staff. Yet only 11% report that more than half their workforce has had any AI-related training.

So the constraint is not access. It is capability, distribution and design: the unglamorous work of making a tool part of how a team actually operates. That work is organisational, and adoption programmes routinely underfund it.

Why AI adoption is a people problem, not a technology one

This sits at the centre of the Future Prep book AI Is Not a Technology Project: AI adoption stalls as a management problem, not a technology failure. The ONS numbers read like an evidence base for the claim. The technology spread fast and cheap. Organisational depth did not follow, because depth was never a procurement line. It turned on skills, process ownership and how work gets redesigned around a capability.

For anyone accountable for AI in an EU organisation, that reframes two things at once. It makes the AI adoption dashboard suspect, since breadth flatters while depth is what governs both risk and return. It also throws the AI literacy duty under Article 4 of the AI Act into sharper relief. A workforce that touches AI without the training to use it well is a strategy gap and a compliance gap in the same breath. The shallow adoption that caps value also leaves obligations half-met.

What to measure instead of reach

Better questions start the fix. Not how many teams have access, but how many run a core process on AI with owners, controls and review. Not whether the firm uses a model, but whether the people around it can challenge its output. Depth over reach. The organisations that crossed cleanly from pilots into production, a move we traced in the governance debt of scaling AI, share one habit. They treat AI adoption as a change in how work is done, and they resource it that way. The ones counting logins keep reporting progress the operations never feel. When productivity gains turned into more work, not less, the lesson ran the same way: the tool is the smallest part of the change.

The uncomfortable reading of the 20 July figures is that most AI adoption is still theatre. Real tools, real logins, shallow change. The organisations that pull ahead will treat the depth gap as the actual project. That case, and how to run AI as the organisational change it really is, is the argument of the Future Prep book AI Is Not a Technology Project.