UK data and AI skills gap 2026: what the numbers say

Part of our topic guides on AI Skills for Business and Data Literacy.

By James Cotton, Founder, iO-Sphere

What the official figures show

Around 35% of UK businesses with 10 or more employees report using at least one AI technology. That comes from the Office for National Statistics, and it is the figure to quote, because it says who was counted and when.

The fieldwork ran from 5 to 28 June 2026, and the earliest comparable reading, from late 2023, was around 12% (ONS, Artificial intelligence in UK businesses: 2023 to 2026, published 20 July 2026).

Adoption rises with company size. In the same release, 28% of businesses with 0 to 9 employees reported using at least one AI technology, against 49% of those with 250 or more.

The spread across sectors is wider still. Over half of businesses in information and communication use AI, at 58%, while construction sits at 13%.

Use tends to be shallow where it exists. Among businesses that use AI, only 10% report using it extensively.

Separately, 15% of all businesses with 10 or more employees report that more than half of their employees use AI as part of their daily work.

Those two figures count different groups. The 10% is a share of the businesses that have adopted AI; the 15% is a share of every business with 10 or more employees.

The technologies named most often are text and image tools. Large language models lead at 18%, then visual content creation at 16%, data processing using machine learning at 12%, image processing at 6% and robotics at 2%.

The training gap

Only 11% of businesses with 10 or more employees report that more than half of their workforce has received AI-related training (ONS, fieldwork 5 to 28 June 2026). That sits under adoption of around 35% from the same survey.

Where businesses do build AI skills, they build them in the staff they already have. Training or retraining existing staff is the most commonly reported route, at around 40% of medium and large businesses.

Recruiting AI-skilled people is rare by comparison, running from about 2% at the smallest businesses to 10% at those with 250 or more employees.

An earlier survey of employers adds detail. In fieldwork run from 19 March to 7 June 2024, 11% of employers said staff had undertaken training on AI in the previous 12 months.

In the same fieldwork, 61% of employers had no staff working with AI at all (DSIT and Ipsos, AI skills for life and work, published 28 January 2026).

Those two 11% figures are different measures. The ONS one counts businesses that have trained more than half the workforce. The DSIT one counts employers with any AI training at all in a year, and it left out sole traders.

Among employers using AI or planning to, 56% rated their organisation's AI knowledge as beginner or novice, in the same 2024 fieldwork.

The barrier employers named most often was not knowing what training would be relevant to the business, at 50%, ahead of time at 47% and cost at 41% (DSIT and Ipsos, 2024 fieldwork). Our guide to AI adoption training for non-technical teams works through that question.

People are ahead of their employers

55% of employees report using AI for work or education, in the ONS Opinions and Lifestyle Survey with fieldwork in May to June 2026.

Set that beside around 35% of businesses with 10 or more employees reporting AI use, and resist subtracting one from the other. These are two surveys of two different populations: one asks businesses what the business does, the other asks people what they do.

Both figures are self-reported, and neither measures how well AI is being used. Data literacy for managers covers what that means for the people supervising the work.

The wider skills-shortage picture

The government's broadest measure of skills shortages is the Employer Skills Survey. The 2024 edition covered 22,712 UK sites, with fieldwork from 24 June 2024 to 31 January 2025, published on 27 November 2025 (DfE, Employer Skills Survey 2024).

Skill-shortage vacancies fell. There were 250,500 of them, down from 531,200 in 2022, and they made up 27% of all vacancies, down from 36%. On the headline measure, hard-to-fill hiring eased.

What changed was their composition. Among skill-shortage vacancies, digital skills were lacking in 38% (32% in 2022), advanced or specialist IT skills in 25% (17%), and complex analytical skills in 50% (43%).

Complex analytical is broader than data and includes problem-solving, so it should not be read as a data-skills figure.

What the survey shows is a smaller pool of hard-to-fill vacancies in which technical and analytical skills are more often the missing part. It does not show the gap widening overall.

The 2024 edition measured AI for the first time. A lack of AI skills was behind 5% of digital skills gaps, among employers who reported digital skills gaps.

There is no current official measure of the UK data skills gap itself.

What the gap is made of

The figures describe use running ahead of training. They do not say what the missing capability is, and that is what a budget has to name.

What training has to build is the ability to assess what a tool has produced: the judgement and habits that come from understanding how AI and data behave and how they fail, matched to what is at stake in the task.

Software multiplies how much work you can produce; it does not supply the judgement that tells you the work is safe to act on. Add the first without the second and you get more output, faster, with nobody able to say whether it is right.

Take an example. A supplier-risk analyst asks an AI tool to summarise a long due-diligence pack. The summary reads cleanly and says the supplier passed its financial checks.

The analyst has read hundreds of these packs. The figures, she notices, are the parent company's, and the parent company is healthy; the entity the contract would be with is losing money. Nothing in the tool flagged the swap.

The catch came from knowing the work. That is the capability in short supply, and it is why it is best built in people who already understand your business.

One official survey put the question to the public as one of confidence. 34% of men and 21% of women in the UK said they felt confident judging whether an AI output was true (DSIT and Ipsos, general public survey, fieldwork 29 February to 7 March 2024).

That is people rating themselves. It came from 1,189 adults, so the gap between the two groups is a rough guide. The AI skills employers need sets out what the capability looks like role by role.

What closes it

The ONS numbers already say what employers do when they need AI skills: they train the people they have. There are good reasons for that.

Capability built into an employee who knows the business stays there and compounds. A contractor leaves and takes the knowledge with them. It also protects money already committed, because an AI licence only returns something once somebody trusts its output enough to act on it.

It outlasts the tools too, since the products your team uses will turn over while the ability to check what they produce carries across. Building novel models is a recruitment problem, and redesigning a whole organisation is consultancy work that iO-Sphere does not do.

Upskilling staff in data and AI covers how that is done in practice.

The Growth and Skills Levy funds apprenticeships for existing employed staff of any age, and for an apprentice under 25 at the start the training costs the employer nothing. What a data analyst apprentice costs an employer and how the levy works are set out separately.

For employed staff, the funded route is an apprenticeship with iO-Sphere as the training provider. For people who are not eligible for one, or who need to move sooner, there is team training in data and AI fluency.

Frequently asked questions

Why do UK AI adoption figures differ so much?

Because they measure different populations, in different ways, at different times. The DfE Employer Skills Survey found 14% of UK sites with two or more staff using AI (broad definition, fieldwork June 2024 to January 2025). DSIT and Ipsos found 31% of employers (excluding sole traders, fieldwork March to June 2024). The ONS finds around 35% of businesses with 10 or more employees (a named list of technologies, fieldwork June 2026). ONS itself says estimates vary substantially with the measurement approach, so quote one figure with its base and date and never line the three up as a trend.

Is there an official measure of the UK data skills gap?

No current one. The last dedicated study is DCMS's Quantifying the UK data skills gap, published on 18 May 2021, and it is too old to present as today's position. What exists instead are adjacent measures: the ONS figures on AI adoption and AI training in businesses with 10 or more employees, the DfE Employer Skills Survey's breakdown of which skills are lacking in shortage vacancies, and the DSIT employer survey on AI skills and training. If a board paper needs a single number for the data skills gap, there is not one to quote, and the case has to be built from the measures that do exist.

Which sectors and sizes are furthest behind?

By sector, construction, at 13% reporting AI use against 58% in information and communication; by size, the smallest firms, at 28% of businesses with 0 to 9 employees against 49% of those with 250 or more (both ONS, fieldwork 5 to 28 June 2026). The DfE Employer Skills Survey 2024, on a different base and an earlier date, found the same two sectors at the extremes (43% and 7%). Adoption and capability are separate questions, and the ONS training figure of 11% is published only as a headline for businesses with 10 or more employees, not by size or sector.

What share of UK jobs will AI touch?

Around 70% of UK workers are in occupations containing tasks that AI could potentially perform or enhance (DSIT and the AI Security Institute, published 28 January 2026). That is modelled exposure calculated from an established index, not a forecast of job losses and not a count of jobs that have changed. It says where AI could reach if it were applied to the tasks it is capable of, which makes it a map of where the question arises rather than an answer to it.