UK AI Use Has Nearly Tripled Since 2023, and the Gap Between Adopters and Everyone Else Is Now a Competitive One

The Office for National Statistics published its analysis of artificial intelligence in UK businesses on 20 July 2026, and the headline number is unusually clear for official statistics. Among businesses with 10 or more employees, self-reported AI use rose from around 12 percent in September 2023 to around 35 percent by June 2026. Adoption is heavily skewed by size: 49 percent of large businesses with 250 or more employees use AI, against 28 percent of businesses with fewer than 10 employees. By sector the spread is wider still, from 58 percent in information and communication down to 13 percent in construction.

Three years ago, using AI was a differentiator. On these numbers it is becoming the default among larger firms while a majority of small ones have not started. That is the gap worth paying attention to, because it is no longer about curiosity. It is about cost per unit of work.

What does the ONS data actually show?

That adoption roughly tripled in under three years, and that it is concentrated. The most reported use across every size band is improving business operations, cited by around 60 percent of adopters, followed by personalising products and services and developing new offerings.

The barriers are just as informative. Around 59 percent of businesses reported challenges with adoption. Lack of expertise was the most cited, at roughly 18 percent and highest among medium-sized firms, with cost between 7 and 14 percent across the size bands. Difficulty identifying a use case and regulatory uncertainty also feature. Notably, 40 percent said their main response to the skills gap is training or retraining existing staff rather than hiring.

Why is lack of expertise the biggest barrier rather than cost?

Because the tools are cheap and the judgment is not. A small business can access capable AI systems for the price of a monthly phone contract. What it cannot buy off the shelf is somebody who knows which of its processes are worth automating, in what order, and where automation would quietly make things worse.

That is why so many small firms report having tried AI without much result. They applied it to whatever was most visible rather than whatever was most repetitive, and got a small saving on a task that was never the constraint.

Where should a small business start?

With the task that happens most often and requires the least judgment. In most small businesses that is not content production, which is where people usually start. It is the handling of enquiries: acknowledging them, qualifying them, answering the same six questions, booking the call, and following up when the prospect goes quiet.

Enquiry handling is high-frequency, rules-based, and directly connected to revenue, which makes it the highest-return first automation for most firms. It is also the area where failure is most visible to a customer, so the improvement is immediately felt. The practical patterns are set out in our piece on AI automation for small businesses.

What should not be automated?

Anything where being wrong is expensive and being slightly slower is not. Pricing decisions, dispute handling, anything requiring a judgment about a specific client's circumstances, and any communication where the customer needs to feel a person is accountable.

The useful test is to ask what happens when the system is confidently wrong. If the answer is a mildly awkward email, automate it. If the answer is a lost client or a regulatory problem, keep a person in the loop and use the system to prepare their work rather than to replace it.

Does this apply to professional firms as well?

Particularly to them. Accounting and professional services sit in the middle of the adoption range, and their work has an unusually high proportion of repeatable, rules-based steps around a core of genuine judgment. The judgment is the product. The repetition around it is overhead.

The firms getting real value are automating the overhead: chasing records, triaging inbound queries, drafting routine correspondence, keeping the client contact rhythm going. That leaves more of the week for the judgment work that clients actually pay for. We covered the specific tools in our roundup of free AI tools UK accountants are using, and the wider commercial framework in our complete guide to digital marketing for accounting firms.

A worked example

A twelve-person firm receives around forty enquiries a month through its website and phone. Each one takes roughly fifteen minutes of somebody's time to acknowledge, qualify and either book or decline. That is ten hours a month, spread across people whose time is worth considerably more than the task.

Automating the acknowledgement, the qualifying questions and the booking link, with a person reviewing anything unusual, cuts that to two or three hours and removes the delay entirely. Response time drops from a day to a minute, which by itself changes conversion. Nothing about the firm's expertise changed. The overhead around it did.

How do you know whether an automation actually worked?

By measuring the thing it was supposed to change, before and after, and by being honest when the answer is nothing. The two numbers worth tracking for enquiry automation are median response time and the proportion of enquiries that receive a second contact. Both are easy to measure and both correlate closely with revenue.

What is not worth tracking is time saved as an abstract figure. Hours saved only count if they were reallocated to something that produces value, and in plenty of small businesses they are simply absorbed. If a four-hour saving did not turn into more client work, more selling or fewer late evenings, the automation was a technical success and a commercial nothing.

The same applies to the quality side. An automated response that arrives in one minute and answers the wrong question is worse than a considered reply the next morning. Sample the output every week for the first month. Most of the failures in AI automation are not dramatic, they are small and steady and nobody looks.

How does this connect to being found in the first place?

Automation improves what happens after somebody makes contact. It does nothing about whether anyone makes contact at all. A firm with excellent enquiry handling and no visibility is efficiently processing a very small number of enquiries.

The sequence that works is visibility first, then conversion, then automation of the follow-up. Being findable is search engine optimisation. Turning a visitor into an enquiry is website design and development. Handling the enquiry consistently without adding headcount is AI automation. The full framework for firms in this market sits on our page for digital marketing for UK businesses.

Why is the size gap widening rather than closing?

Because adoption compounds. A large business that started in 2023 has three years of internal knowledge about which processes suit automation, which vendors are reliable, and how to roll a change out without breaking anything. That knowledge is worth more than the tools, and it cannot be bought later at the same price.

The encouraging part of the ONS picture is that the gap is one of practice rather than capital. At 28 percent, businesses with fewer than 10 employees are behind, but the barrier they report most is expertise, not cost. Expertise is closable in weeks by a small firm that picks one process and finishes it. That is a very different kind of disadvantage from one that requires a budget the business does not have.

What about the training point?

The ONS finding that 40 percent of businesses are training existing staff rather than hiring is the most practical signal in the release. It matches what actually works. The person who already knows why your process exists in its current form is far better placed to automate it sensibly than a new hire who knows the tools but not the business.

For a small firm that means picking one person, giving them a defined amount of time, and letting them automate one process end to end rather than dabbling across five. One completed automation that saves four hours a week is worth more than five half-built ones that save nothing and have to be maintained.

What to do next

Count how many times your business performed its single most repetitive task last month, and multiply by the minutes it takes. That number tells you whether automation is worth doing at all, and it is usually larger than owners expect.

Then start with enquiry handling, keep a person accountable for anything with judgment in it, and measure response time before and after. If your response time to a new enquiry is currently measured in hours, that is the cheapest competitive gain available to you this quarter.

Triomatic Marketing builds visibility, conversion and automation systems for businesses across the UK, USA and Pakistan. We are AI-powered and founder-led, and we sequence the work so the automation has something to work on. To talk it through, message Aria on WhatsApp via triomaticmarketing.com, or book a free 15-minute discovery call at https://calendly.com/hello-triomaticmarketing/15min.

Frequently asked questions

FAQs