AI skills gap: why two thirds of UK firms using AI say they lack the know-how
· 5 min read
The UK government's AI adoption survey 2026 finds adoption has doubled but skills are the biggest barrier, even for firms already using AI. What it means for leaders and which toolkit tools help.
In brief
The UK government's AI adoption survey for 2026 shows that the share of UK businesses using AI has doubled in a year, to roughly one in three. Yet the main thing slowing them down is not price, regulation or the technology. It is capability. Two thirds of businesses that already use AI say limited skills, expertise or knowledge holds them back, and the figure is higher still among firms that plan to start. Many firms not yet using AI also see no clear business case. Productivity gains are real but mostly modest, most firms have seen no immediate change in operating costs, and the share keeping a person in charge of checking AI output has slipped a little since last year. This article looks at what those findings mean for leaders. It argues that an AI skills gap is rarely one gap, because people sit at very different stages, and that the quality of human oversight depends on whether the checker knows what good looks like. Drawing on a principle from John Sansbury's book about frameworks that explain what to do but not how, it sets out which tools in the AI Enablement Toolkit for Leaders help teams build their own practical skills, where Esm training for managers fits, and what the toolkit cannot do.
One in three UK businesses now uses at least one AI technology. A year ago it was one in six. On the face of it, that is a story about momentum.
Read the UK government's AI adoption survey 2026 a little further, though, and a different story appears. The biggest thing holding businesses back is not money, regulation or the technology itself. It is knowing how to use it.
What the survey found
The survey was published by the UK government on 29 October 2026. It is based on 3,503 telephone interviews with businesses carried out between 7 January and 21 April 2026, followed by 50 in-depth interviews and a set of case studies. These are the findings that matter most for leaders:
- Adoption doubled, from 16% of businesses in 2025 to 32% in 2026, and 52% among larger firms.
- Limited AI skills, expertise or knowledge affected 67% of businesses that had already adopted AI, and 78% of those planning to.
- 65% of non-adopters cited the lack of a clear business case or proven return on investment.
- 56% of adopters, and 60% of those planning to adopt, had concerns about bias and lack of accuracy.
- Half of adopters reported higher productivity, but the gains were generally described as modest. 64% had seen no immediate effect on operating costs.
- Only 14% considered AI tools to be of high or critical importance to their operations.
- 81% apply some level of human input or oversight to AI output, a figure the report says has declined slightly since 2025.
The report also notes that cost, where it was a barrier, was more about the time and effort needed to implement and embed AI than the price of the tools.
Two other reports published the same day point the same way. Research by The Productivity Institute, commissioned by the technology reseller Bytes Technology Group, found that 89% of UK organisations use AI somewhere but only one in four has changed how work is done to make the most of it. Deloitte's latest State of AI in the Enterprise found that 60% of companies have redesigned jobs around AI to some degree, but only 15% report a significant shift in who makes decisions. Both are commercial studies and should be read as indicative.
What it means for leaders
The striking number is the 67%. These are not sceptics standing on the sidelines. They are businesses that have bought the tools, and they are telling us they do not yet have the skills to use them well.
That gap is easy to misread. The reflex is to buy a course, roll it out to everyone, and count completions. But a skills gap in AI is rarely one gap. Some people have barely tried the tools. Some are capable but have quietly stopped. A few are fluent and wondering why the process around them has not changed. The same training lands differently on each of them.
There is a second risk hiding in the oversight figure. If 81% of firms keep a human check on AI output, the value of that check depends entirely on whether the person doing it knows what good looks like. A skills gap and an oversight gap are often the same gap.
John Sansbury describes the same problem in service management, long before generative AI, in The Journey to Service Excellence:
"There are many service management frameworks; however, they all have one major limitation: they describe the ‘what’ but not the ‘how.’"
Applied to AI, that is Esm's reading rather than John's: the tools arrive full of "what", features, prompts and possibilities. The "how", what good use looks like in this team, on this work, for this customer, has to be built by the organisation itself. That is leadership work.
Which toolkit tools help
The AI Enablement Toolkit for Leaders is a leadership method, not software. Several of its 26 tools speak directly to what this survey found.
- Tool 4, the naive-to-native adoption ladder, helps leaders see where each person or team actually is, so support fits the stage rather than giving everyone the same training.
- Tool 7, see the work, asks the people doing the job what they know that the system does not. Those exceptions and workarounds are where good AI use starts.
- Tool 13, the incentive and behaviour audit, includes a "capable stall" test: if people know how to use AI and still revert to the old way, inspect the workflow and reward system before buying more training.
- Tool 5, the value case canvas, answers the 65% of non-adopters who see no business case, by starting from a customer or business outcome rather than a tool.
- Tool 6, the output-to-outcome measure reset, stops usage and logins being mistaken for results.
Where training fits
The survey points to a capability gap, and capability is built by people as much as by documents. This is where Esm training and enablement helps most: equipping managers to diagnose where their teams are, to teach the "how" in their own work, and to design oversight that someone is actually qualified to carry out. Training for staff on the tools themselves still matters, but it works best when leaders have first decided what the tools are for and what good looks like.
Honest limits
The government survey is self-reported, and its fieldwork finished in April 2026, so the picture may already have moved. The toolkit is a management method. It does not teach people to use particular AI products, fix data quality, or replace legal, data protection or sector-specific advice where AI touches regulated decisions. It also has no tool for tracking AI running costs, which needs finance and technical input.
What it can do is help a leadership team ask a better question than "have people been trained?" The better question is whether anyone in the organisation owns building the how, and whether they would recognise it when they saw it.
Tags: ai-adoption, skills, leadership, uk-business, training
Sources
Join the Exchange free
Join free to unlock the complete knowledge library.
Founder membership is free for life
More from Esm
How do I find everyday AI uses for staff outside IT?
5 min read
Experience debt: what happens to judgement when AI takes the tasks people learned from
5 min read
How should IT teams handle emergency patches for critical vulnerabilities?
5 min read
When the human in the loop becomes a rubber stamp: what UK regulators found about AI checks
5 min read
10 scheduled AI tasks for service management teams
8 min read
Boards expect AI savings: why almost no organisation is finding them
5 min read
Related articles
How do I find everyday AI uses for staff outside IT?
Start with the work your teams already do, not a list of AI tricks: observe, set data rules, run small timed trials with an owner, and publish what works as knowledge.
Experience debt: what happens to judgement when AI takes the tasks people learned from
New research finds most people say skills are slipping as AI takes on routine tasks, while time to learn is scarce. What that means for leaders, and which toolkit tools help rebuild the learning route.
How should IT teams handle emergency patches for critical vulnerabilities?
Treat an emergency patch as an emergency change: triage the risk fast, agree the fast path in advance, test as far as the clock allows and keep a remediation plan.
When the human in the loop becomes a rubber stamp: what UK regulators found about AI checks
The Bank of England and FCA find most firms approve AI before launch but far fewer keep watching it, and the FCA warns human review can become a tick-box exercise. What leaders should check, and which toolkit tools help.
Boards expect AI savings: why almost no organisation is finding them
Gartner finds 77% of board members expect AI to deliver savings, but only 3% of organisations beat their savings goals through productivity. What leaders should agree instead, and which toolkit tools help.
AI rebounding: why some employers are asking back the people AI replaced
New SHL research finds some UK and US workers replaced by AI were later asked back. Why it happens, and how leaders can understand the work before automating it.