ONS: a third of UK businesses use AI, and one in ten use it seriously

What changed

The Office for National Statistics published a new analytical article, "Artificial intelligence in UK businesses: 2023 to 2026", on 20 July 2026. It draws on the Business Insights and Conditions Survey, and for the first time it asks not just whether businesses use AI but how deeply. Among businesses with 10 or more employees, use of at least one AI technology has risen from around 12 per cent in September 2023 to around 35 per cent in June 2026. Across all size bands the figure is 29 per cent. The depth questions are the new part, and they tell a different story: the average adopting business uses 1.6 AI technologies, up from 1.4 in 2023, and only 10 per cent of adopters describe their use as extensive. ONS's own conclusion is that this "would imply relatively limited transformative impacts to date for most AI-adopting firms".

Why it matters for UK business

Most AI adoption statistics measure whether a business has started. This one measures whether it has got anywhere, and the gap between the two numbers is the useful finding. A third of firms have adopted something. One in ten of those is using it seriously. That is not a story about laggards failing to begin; it is a story about a large group of firms that began and then stopped moving.

The training figures point at why. Only 11 per cent of businesses with 10 or more employees say more than half of their workforce has received AI-related training. A licence issued is not a capability acquired, and an average of 1.6 technologies per adopter is roughly what you would expect from a firm that bought a chat subscription, found it useful, and never went further. That is a reasonable place to be after a year. It is a poor place to still be after three.

For a UK SME the practical read is that the competitive benchmark is lower than the noise suggests, and the bar for being genuinely ahead is lower still. If a third of your peers have adopted and one in ten of those has depth, then serious use puts a firm in a small minority. That is a more accurate picture than either the vendor-led narrative of universal transformation or the reflexive counter-narrative that none of it works.

Two honest qualifications belong on this data. It is a voluntary survey with a 26.7 per cent response rate, and ONS classes the outputs as official statistics in development, so the depth questions in particular carry one wave of data and ONS asks that they be read as early indications. It also excludes finance and insurance and the public provision of health and education, which removes two sectors where a good deal of UK AI work is happening. And ONS itself flags that adoption estimates vary substantially by how you ask: it cites separate research reporting around 71 per cent of UK firms when senior executives are surveyed directly, against its own lower bottom-up measure.

There is one finding that cuts against the way consultancies usually present data like this, and it is worth stating rather than burying. Among firms that cite a lack of AI expertise as a barrier, around 62 per cent respond by training or retraining existing staff, and ONS notes that outsourcing remains relatively uncommon. A further 41 per cent of businesses report that nothing prevented or delayed their AI adoption in the last three months. This is not a picture of widespread frustrated demand for outside help.

What to do, and what not to do

Do:

• Benchmark yourself on depth, not adoption. Count how many AI technologies your business genuinely uses in production, how many people use them weekly, and what proportion of the workforce has had any training. Those three numbers place you against this data honestly.
• Treat the 1.6 average as a diagnostic. If your firm has adopted one tool and stopped, the constraint is usually capability or process rather than product, and buying a second tool will not fix it.
• Put training in front of tooling if you have to choose. The 11 per cent training figure is the most actionable number in the release, because it is both low and directly within a business's control.
• Use the sector spread when setting expectations. Adoption runs from 58 per cent in information and communication to 13 per cent in construction, so a sensible target depends heavily on which of those your business looks like.

Do not:

• Quote "35 per cent of UK businesses use AI" without the size qualifier. That figure covers businesses with 10 or more employees; the all-size figure is 29 per cent and micro businesses dominate the UK business population.
• Read the 10 per cent extensive-use figure as a trend. It is one wave of a newly introduced question and ONS explicitly asks that it be treated as an early indication.
• Treat this as evidence that AI is not working. ONS is measuring breadth of technologies, not returns, and explicitly says its headline measure "treats all reported uses as equivalent".
• Assume the answer is external help. The ONS data says the dominant response to a capability gap is training existing staff, and that is often the right answer.

Where The AI Consultancy fits

The honest reading of this data is that most firms who have stalled need capability rather than procurement, and ONS finds most of them build that capability internally. Where we are useful is narrower than the headline invites: setting the direction, deciding which two or three workflows justify real depth, and giving an internal team enough structure to keep going without a permanent external dependency. That is the shape of our fractional AI officer engagements. If you would rather test your own position first, the AI readiness assessment covers it, and our briefing on why agentic AI projects stall makes the same argument from the delivery side.

All figures are drawn from the ONS analytical article "Artificial intelligence in UK businesses: 2023 to 2026" (20 July 2026) and the Business Insights and Conditions Survey bulletin of 2 July 2026, covering Wave 159, fieldwork 15 to 28 June 2026. BICS is a voluntary survey; its outputs are official statistics in development and carry an ONS uncertainty caution.