UK small businesses and AI: the gap isn't adoption, it's depth

13 October 2026Andrew Cheeseman

Walk down any high street in Sussex and you can see the problem in the shop windows. Some are lit and busy. Some are dark. From the pavement you can't tell which businesses are quietly using AI and which aren't. The official numbers can.

What the numbers say

The Office for National Statistics (Business Insights and Conditions Survey, fieldwork June 2026) found that 28% of businesses with fewer than ten employees use at least one AI technology, against 35% for firms with ten or more and 49% for those with 250 or more. For firms with ten or more staff, that is up from around 12% in late 2023. Adoption has roughly tripled in under three years.

A separate government survey, the UK Business Data Survey 2026 (fieldwork October 2025 to January 2026), puts the figures higher: 41% of micro businesses and 51% of small firms. The two aren't contradictory. They measure different things: that survey covers only businesses that handle digitised data and asks about AI use for any purpose. Treat the ONS as the more cautious reading and the DSIT-published survey as the more generous one. Either way the shape is the same: the smaller you are, the less likely you are to be using it.

The more important number

Adoption headlines hide what matters. Among adopters with ten or more employees, the ONS found:

They use on average 1.6 AI technologies, barely up from 1.4 in 2023.
Only 10% describe their use as extensive.
Only 15% say more than half their staff use AI in daily work.
Only 11% have given AI training to more than half their workforce.

So the picture isn't "most firms have AI and a few are behind". It's "a lot of firms have tried a chatbot, and very few have changed how the business works". That is where the gap really is.

What small firms are actually doing

By far the most common technology is the large language model, the chat assistant. Of firms with ten or more staff, 18% use them, followed by AI image and content creation at 16%. That fits what I see: writing emails, summarising documents, drafting marketing copy. Useful, cheap, and individual. It is a personal productivity tool, not a business process.

By sector it is lopsided. Information and communication firms are at 58% and professional, scientific and technical at 46%; construction is at 13%. A small architect's practice in Hove and a family building firm in Worthing are in very different places.

The consequences of standing still

I want to be careful here, because the data shows adoption rates, not outcomes, and I'd be making things up if I claimed a precise cost of not adopting. What follows is my professional judgement, not a statistic.

Your competitors' admin costs fall and yours don't. Quoting, invoicing, scheduling, customer replies: if a rival turns a four-hour task into forty minutes, they can price lower or answer faster.
Your people use it anyway. If you have no policy, staff will paste customer data into free tools on their own phones. That is a data-protection risk (UK GDPR) you've taken on without choosing it.
The skills gap compounds. Training is the most common way firms build AI skills, and firms that already report a lack of expertise are far more likely to be training (about 62%) than those reporting no barrier (about 26%). The firms that learn first learn faster.
Suppliers and customers will expect it. Larger customers are moving; the ones who deal with you will start asking how you handle data, quoting and response times.

A sensible first ninety days

None of this requires a big programme. For a firm of 5 to 50 people:

Pick one painful, repetitive process, not "AI strategy". Quoting, enquiry triage, month-end reporting, rota changes.
Write a one-page policy. What can and can't go into AI tools, which tools are approved, who decides. Half a page on customer data is enough to start.
Train everyone a little, not one person a lot. The ONS numbers show broad training is rare. A short session on real tasks beats a licence nobody uses.
Measure one thing. Time saved, response time, error rate. If it doesn't move, stop.
Keep a human accountable. AI drafts; a named person signs off.

I'd add one caution from my own work: AI amplifies the processes you already have. If your quoting process is chaotic, AI makes the chaos faster. Fix the process, then automate it.

Where I can help

I work with small and mid-sized UK firms as a fractional chief architect: working out which processes are worth automating, how to connect your existing systems safely, and how to build your team's skills so you don't depend on me. A free 30-minute call is the quickest way to find out whether I'm the right person: cheesesharp.com