UK AI adoption has doubled in a year. What has not changed?

Adoption is the easiest thing to measure about AI, and often the least revealing. The government's latest survey of UK businesses shows why. In 2026, 32% of businesses with five or more employees were using at least one AI technology, up from 16% a year earlier. That is a doubling in twelve months, and it will rightly make headlines.

However, when I put the 2026 report next to its 2025 predecessor, the more interesting story sits underneath that number. Has AI become wider, deeper and more valuable inside UK business? The honest answer is yes, not yet and not so far. I will take the three questions in turn, set the evidence side by side, and then say what I think organisations should do about it.

Wider: more businesses, in every corner

The first answer is yes. Both surveys used the same basic approach: around 3,500 telephone interviews with private sector businesses, speaking to the person responsible for technology. Between them, the share of businesses saying AI was not relevant to their organisation fell from 51% to 38%. Adoption rose in every size band. Large businesses moved from 36% to 52%, medium-sized from 23% to 47%, small from 16% to 34% and micro businesses from 14% to 27%.

The sectors that started furthest behind moved fastest in relative terms. Construction went from 9% to 25%, retail and distribution from 10% to 25%, and transport and storage from 7% to 24%. Information and communication, already the leader, went from 43% to 61%. However, the gaps have not closed. The distance between information and communication and construction was 34 percentage points in 2025 and 36 in 2026. Everyone is moving, but the order has not changed.

Why now? What has changed in the way businesses talk about AI? Efficiency is still the leading motive, cited by 65% of businesses using or planning to use AI in 2025 and 71% in 2026. But gaining a competitive advantage or keeping up with competitors doubled, from 12% to 25%, and improving customer experience rose from 16% to 23%. As one small finance business in the West Midlands told the researchers, "if there are competitors out there who can run more streamlined, we're going to be at a loss". The 2026 report itself says that adoption is increasingly driven by the need to keep pace with competitors.

Deeper: not yet

Here the picture is different. Among businesses using AI, the average share of staff using it was 30% in 2025 and 31% in 2026. The proportion of adopters with more than half of their staff using AI barely moved, from 21% to 22%. Indeed, twice as many firms are using AI, but the layer of use inside each of them is just as thin. Constant use rose a little, from 53% to 57% of adopters.

The technology mix tells the same story. In 2025, 85% of adopters were using natural language processing and text generation, and agentic AI was the least used technology, at 7%. In 2026 the report again finds use concentrated in accessible, low-complexity tools, with agentic AI still used by 7% of adopters and OpenAI models used by 92% of them. I think about four modes of using generative AI: sophisticated search, content creation, thought partner and agent. These surveys suggest that most UK businesses are still operating in the first two.

There are signs of movement at the edges. Among users of each technology, the share that developed it in-house, or by adapting existing software, rose for machine learning (24% to 45%), agentic AI (12% to 34%) and AI-related hardware (17% to 31%). These are small groups, with only 88 agentic AI users in the 2026 sample and 59 in 2025, so I would read them as a signal and not a trend.

Then there are the intensive adopters. The 2026 report defines them as businesses that build bespoke AI or embed it in existing processes, rate it as highly important to the organisation and use it constantly. They make up just 2% of all businesses surveyed. Their profile is different: on average 60% of their staff use AI, against 31% for adopters overall, and they spent an average of £31,300 on AI in 2025 compared with £7,800.

More valuable: modest so far

On value, the 2026 survey reads like a careful accountant. Among adopters, 71% reported improved workforce productivity (75% in 2025) and 60% reported more efficient or flexible processes (57%). Yet only half (50%) reported an overall increase in employee productivity, down from 56%, and the report describes the gains as modest. On costs, 24% reported lower operating costs, 64% saw no immediate effect and 9% saw an increase. The 2025 survey asked about revenue instead, and 77% of adopters had not seen a change.

Perhaps the most telling figure is about importance. Asked to rate how important their AI tools are to their operations on a scale of 0 to 10, only 14% of adopters gave 8 or higher, while 44% gave 3 or lower.

Spending points the same way. Among businesses using or planning to use AI that could estimate their spend, 47% said they spent nothing on AI in 2025, up from 31% for 2024 in the first survey. The mean spend fell from £19,000 to £7,500. The report attributes the zero-spend group to free tools and AI built into existing software. My own reading is that a wave of small first-time adopters has pulled the average down, which is a good reminder of why averages mislead.

The 2% who have gone furthest look different. Among the intensive adopters, 81% reported higher employee productivity (against 50% of all adopters) and 54% reported lower operating costs (against 24%). They also more often reported higher costs (20% against 9%). Cause and effect are not proven, since these may simply be better resourced or better managed firms, and all of the figures are self-reported. However, the pattern suggests that value follows depth, and that depth has to be paid for.

Wide, deep, valuable: the two surveys side by side

Question	Measure	2025	2026
Wide	Businesses using AI	16%	32%
Wide	Businesses saying AI is not relevant	51%	38%
Wide	Large businesses using AI (250+ employees)	36%	52%
Wide	Micro businesses using AI (5 to 9 employees)	14%	27%
Deep	Average share of staff using AI, among adopters	30%	31%
Deep	Adopters with more than half of staff using AI	21%	22%
Deep	Adopters using AI constantly	53%	57%
Deep	Adopters using agentic AI	7%	7%
Valuable	Adopters reporting improved workforce productivity	75%	71%
Valuable	Adopters reporting an overall productivity increase	56%	50%

Breadth has doubled, depth has barely moved, and the value that businesses report has slipped slightly. My reading is that a wave of newer adopters starting with the easy tools would produce exactly that pattern.

What stands in the way

The barriers have shifted with the adoption curve. In 2025 the reasons most often given for not adopting were a lack of identified need and limited AI skills, and among the businesses that cited them, ethical concerns, high cost and unclear regulation were rated the most significant barriers. By 2026 the group who see AI as irrelevant has shrunk, and the barriers are about capability. Among adopters, 67% cited limited skills or knowledge and 64% limited knowledge of how AI could help the business. Among businesses planning to adopt, 78% cited skills. Among those with no plans, the most common barrier was the lack of a clear business case (65%). Cost was described less as the price of the tools and more as the time and effort needed to embed them. Hence the question is moving from "why would I?" to "how would I?".

Governance has not kept pace. Among adopters, 81% reported at least some human checking of AI outputs, down from 84%, and the proportion reporting none rose from 2% to 5%. The report suggests this may reflect growing trust in outputs. It may. It may equally be early complacency, and a survey cannot tell us which. The interviews indicate that documented policies are still uncommon, with most firms relying on informal practice and several expecting to formalise rules only as use grows.

A caution on the comparison

Two cautions apply. The surveys are comparable by design but not identical. Different agencies ran each wave (IFF Research and Technopolis Group in 2025, BMG Research in 2026), the fieldwork windows differ (February to May 2025 and January to April 2026), and the 2026 interviews were shorter, at 15 minutes against 20. Some questions also changed, which is why I have not compared financial impact, where 2025 asked about revenue and 2026 about operating costs, or the share of adopters using text generation, where the bases differ.

Both surveys also count a business as an adopter if it is piloting AI, and both speak to one person per business, usually the person responsible for technology. The first report notes that this approach cannot capture shadow AI, meaning tools used by staff that the organisation does not know about. That means both surveys may understate use, and neither can tell us by how much.

Three things worth doing

If adoption is no longer the question, what should a leadership team ask instead? Here are three things worth doing.

Measure depth as well as presence. Ask what share of your people use AI, for which tasks, and which of the four modes they are in. The 31% average among adopters is a rough benchmark, but the more telling comparison is with your own figure a year ago. Counting the tools you have bought says very little.

Pick one process and embed AI in it, with a number attached. The intensive adopters differ from the rest because they built or embedded AI in real processes and spent accordingly. Choose a single process, record the hours, errors or cost per transaction before you start, and measure again after three months. That is the cheapest way to build the business case that 65% of businesses with no plans say they lack, and it tackles the skills gap that 67% of adopters report.

Write the rules before the scale arrives. With use still informal and human checking slipping slightly, it is cheaper to decide now who reviews which outputs, which tasks are off limits and how AI-assisted work is recorded than to do so after the first costly mistake. Several of the firms interviewed said they planned to formalise policies only as use grew.

None of this requires a transformation programme. It requires curiosity about what is happening inside your own organisation, which is, after all, what the government's survey is trying to find out about the country.

Sources

• UK Government (Department for Science, Innovation and Technology), Artificial Intelligence Adoption Research, January 2026. Survey of 3,500 UK businesses with five or more employees, fieldwork 12 February to 2 May 2025, conducted by IFF Research and Technopolis Group.

• UK Government, Artificial intelligence adoption research 2026, October 2026. Survey of 3,503 UK businesses, fieldwork 7 January to 21 April 2026, conducted by BMG Research.

• All figures come from these two reports and are self-reported by businesses. Quotations are from the in-depth interviews in the 2026 report.