More users, smaller reported gains
The government’s second AI adoption survey, published on 29 October, found 32% of UK businesses with five or more staff using AI, against 16% a year earlier. Further into the full report, a second comparison runs the other way. Among adopters, the share reporting higher employee productivity fell from 56% to 50%, and the share reporting a gain of 20% or more fell from 16% to 7%. Our news report covers the publication itself. This piece asks what broad, shallow use means for UK productivity plans, because the government has already said the real prize lies in depth.
Strategic Insight: The doubling counts businesses “currently using it (including piloting)”. The OECD range the Budget cites is built on a much narrower idea of adoption: regular, high-intensity use in core business functions. So the 32% cannot simply be read against that range. And on one measure of reach the survey does offer, the average share of staff using AI at adopting firms, the figure barely moved: 30% in 2025, 31% in 2026.
What the productivity case rests on
DSIT, which no longer exists, commissioned the research from BMG Research; the report carries the imprint of the Department for Business, Innovation, Science and Trade, one of three bodies that took on its work. Its results land just after a Budget that used AI to show “what is at stake” for UK productivity.
The published estimates vary widely, and adoption runs through all of them.
| Source | Estimate for UK productivity growth | Stated condition |
|---|---|---|
| OECD, as cited in Budget 2026 | 0.4–1.3 percentage points a year over the next decade | ”depending on the rate of diffusion and adoption” |
| The Productivity Institute, commissioned by Bytes | 0.5–0.7 points a year | Its authors tie this to closing the “AI ambition gap” |
| Office for Budget Responsibility (OBR), October 2026 central forecast | 0.2 points a year by the medium term | ”the scale and timing of any effect is highly uncertain” |
| OBR briefing paper, November 2025 scenarios | Closer to 0.5 points in its upside case; negligible in its conservative case | The conservative case is one “in which adoption is more limited” |
The gap between those numbers has a price. The OBR has illustrated that productivity growth half a point above or below its central forecast could mean borrowing around £50 billion lower or £40 billion higher by 2030-31. That illustration covers productivity in general, not AI alone.
The OECD range does not count what the 32% counts
The Budget’s range comes from a 2025 OECD working paper, which carries its authors’ views rather than the OECD’s. For the UK it runs from 0.39 points in a slow-adoption scenario to 0.97 in a medium one and 1.27 in a rapid one. The upper two also assume expanded AI capabilities, so adoption is not the only thing moving.
The adoption term is the important part. The authors count “high-intensity, regular use of AI in core business functions and operations”, and say this goes beyond “the occasional reliance on AI tools by employees in isolated tasks”. On that definition, their preferred estimate for the UK in 2024 is 5.4% of businesses. The official UK headline they set beside it, for any AI use at firms with ten or more staff, is 26.5%. Their scenarios then take adoption on the narrow definition to between 30% and about 60% by 2034, for a group including the UK.
It would be easy to set the survey’s 32% against that and conclude the UK is already at the low end, a decade early. That would be a category error. The 26.5% and the 5.4% describe the same year and differ almost fivefold, which shows how far apart the two ideas of adoption sit. The 32% includes pilots and occasional use. Nothing in the survey matches the OECD definition exactly, but on our arithmetic the nearer proxies sit far lower: embedding AI into applications and processes is 40% of adopters, or about 13% of in-scope businesses; rating AI highly important is about 4.5%; and intensive adopters are 2%.
The speed of the rise does not rescue the comparison. The authors read a rise of about 50% a year in official EU and US data as most consistent with their fastest path, but those data started from around 3-5%, not the 16% this survey started from.
The 5.4% is also modelled. With no comparable UK survey data, the authors predicted it from digital infrastructure and skills, and they warn that UK estimates carry higher uncertainty than most. So the range the Budget quotes starts, for the UK, from an estimate of current adoption rather than a measurement. In our view, that is the gap this survey series could help close if it reported depth on a comparable basis.
The line the government drew in June
In June, the government’s interim response to the AI Champions’ adoption plans drew the line itself: “The real gains from AI will come from deep adoption”. It defined that as redesigning how businesses work, and contrasted it with firms using AI on existing processes, “such as drafting emails, carrying out research or writing code”.
The survey’s findings sit mostly in that second group. Its interviewees mostly used large language models to cut administrative work, and the report’s summary judgement is that AI is “augmenting existing activities rather than redefining business models”. In our view that is the sentence to read beside any quotation of the 32%.
| Measure | 2025 | 2026 | What it tells us |
|---|---|---|---|
| Businesses using AI, including pilots | 16% | 32% | Breadth doubled |
| Businesses calling AI not relevant | 51% | 38% | Resistance is falling |
| Average share of staff using AI (adopters) | 30% | 31% | Reach inside firms is flat |
| Adopters reporting higher productivity | 56% | 50% | Fewer report any gain |
| Adopters reporting a gain of 20% or more | 16% | 7% | Large gains became rarer |
| Adopters reporting no productivity change | 35% | 42% | The flat result grew |
| Those expecting a bigger AI budget | 71% | 65% | Still most, but fewer |
Strategic Reality: The 14% and the 64% are shares of adopters, not of all firms. The executive summary says “14% of businesses”, but the chapter behind it gives the base as those using AI, and defines high importance as 8 to 10 on a 0 to 10 scale. Multiply 14% by 32% and, on our arithmetic, roughly 4.5% of in-scope businesses rate AI that highly. A further 44% of adopters scored it 0 to 3.
How to read a falling number in a rising market
Cheap adoption is easy to double
Natural language processing and text generation went from 13% to 30% of all in-scope businesses, while machine learning reached 6% and computer vision 9%. Among current users who could put a figure on 2025 spending, 43% said it was nothing, which the report puts down to built-in features or free or low-cost tools. One mid-sized arts business told researchers: “We just use the free ones like ChatGPT”.
None of that makes the use worthless. It does mean the 32% is a weak signal of the investment the gains are said to need. The Productivity Institute’s authors point to management, skills and redesigned workflows. The OECD authors say deeper integration “may still require significant complementary investments”, in data, skills and the reorganisation of business processes.
What depth looks like, and why to be careful
The report isolates a group it calls intensive adopters: firms that embed AI or build bespoke tools, rate it highly important and use it constantly. They number 101 unweighted cases, or 2% of surveyed businesses.
Their results differ sharply. Some 81% report higher employee productivity, against 50% of all adopters. Some 54% report lower operating costs, against 24%. They spent an average of £31,300 on AI in 2025, against £7,800, and their AI use reaches 60% of staff on average, against 31%.
Reality Check: This is not proof that depth pays. The group is partly defined by rating AI as highly important, so it selects firms already pleased with the results, and the sample is small. Intensive adopters were also likelier to report higher operating costs (20% against 9%). The report attaches the same caution to its finding that agentic AI users most often report lower costs, calling it “indicative but not conclusive”.
Three reasons the fall may overstate the problem
First, composition. The report does not test why the productivity share fell. Doubling the adopter base brings in many newer users, and in our view that alone could lower the average without any established user doing worse. The published figures cannot separate the two.
Second, timing. The OBR expects AI’s effect to build gradually “due to adoption lags and the need for complementary investments”, so shallow results in early 2026 do not refute a decade-long estimate. The OECD authors go further: where complementary investment is needed, measured productivity “may respond slowly to adoption, or even show a temporary decline”. The survey’s own fieldwork closed in April, too.
Third, measurement. These are self-reported answers from telephone interviews averaging 15 minutes, with the person responsible for technology. They are not output data. Output data do not settle it either: the OBR finds productivity growth has accelerated in sectors tied to AI, such as IT and communications, but also in sectors it thinks less affected, including accommodation and food.
Critical Context: Official statistics point the same way on depth but differently on pace. The Office for National Statistics (ONS) found adoption “relatively shallow” in July, with 10% of adopters at firms of ten or more staff reporting extensive use. The OBR notes the ONS measure reached 30% in September 2026 from 15% two years earlier. That took twice as long as the survey’s doubling, though the two cover different populations.
Who the gap affects most
The survey’s averages hide very different positions. Where it speaks to a group, the effect is below; where it does not, the question stays open.
| Group | What the evidence shows | What is still unknown |
|---|---|---|
| Treasury and the OBR | Forecasts depend on adoption the survey shows is mostly shallow | Whether shallow use is a phase or a resting point |
| Businesses with 5 to 9 staff | 27% use AI; 52% of those estimating spend reported £0 | Whether free tools lead on to deeper use |
| Hospitality, transport, retail and construction | Use runs at 23–25%, and these firms are likelier to see AI as irrelevant | How much of that reflects low exposure rather than low uptake |
| Legal, finance, healthcare and transport | Regulatory and liability worries are holding adoption back | Whether sector rules ease this |
| Public services | Outside the survey: it excludes public administration, education and health | What the NHS review finds in early 2027 |
What this means for firms setting AI budgets
Planned investment leans towards the same pattern as current use: off-the-shelf applications (61%), task automation (57%) and embedding AI in existing systems (54%). The cost firms describe is mostly the time and effort of implementation, not the tools. One business put the effort of a recent technology rollout at “maybe a quarter to a fifth of our time”.
So in our view, comparing licence counts with competitors tells a board little. The more useful internal question is the one the survey could only ask loosely: what share of staff use AI, on what work, and whether anyone measured the result before and after. Firms that cannot answer that are in the same position as the national statistics.
The questions the survey cannot settle
It counts firms, not output
The survey is weighted to represent businesses, by size and industry. Large firms lead on adoption (52%, against 27% of the smallest), yet the headline is a share of businesses, not of jobs or output. A productivity plan needs to know how much of the economy’s output has changed, and a count of firms cannot say.
Depth has only just started being measured
The ONS added questions on the extent of AI use and the share of employees using it in a June 2026 wave, and says they will help measure intensity more directly in future. The June government response also announced an AI Economics Institute, chaired by Simon Johnson, to follow AI’s effects on employment and growth. The OECD authors say harmonising AI adoption data would be a key step forward. Both date only from this year, too recently to offer a long run of data on depth.
The priority sectors are not reported as such
The report says understanding adoption across the eight Industrial Strategy sectors is particularly important. Neither it nor its data tables present results under that heading: they use broader industry groups. Anyone checking progress against the sector plans has to approximate.
Support is being tested, not scaled
Limited skills or knowledge remains the leading barrier, and readiness has not improved: 53% of adopters feel ready to scale, similar to last year’s 54%, while the share who do not rose from 12% to 19%. The Budget confirms pilots of local AI adoption partnerships, starting with three Yorkshire authorities. Whether such pilots reach the managers who decide how work is organised is unresolved.
What to watch
Four things would show whether the 32% is becoming the productivity gain the plans need.
- The OBR’s next forecast. It says it will keep monitoring the evidence. A move from 0.2 points towards its 0.5 upside would suggest it sees depth arriving; a cut would suggest the conservative scenario has gained weight.
- The ONS extensive-use figure. It stood at 10% in June. If it climbs while headline adoption stays near 30%, firms are going deeper rather than wider. The OECD authors suggest tracking adoption as a short-run gauge, since productivity data arrive late.
- Any third wave of this survey. We have found no published date. Firms have given it something to test: 31% of current and prospective users expect over half their workforce on AI within two years, against 22% today. The figures to compare are that one, the 7% reporting gains of 20% or more and the 14%.
- The NHS productivity and AI review. Led by Mark Bailie and due in early 2027, it covers part of the economy this survey leaves out, and the Budget says it will inform the next spending review.
The report itself sees a path to deeper integration, and it may be right. But the doubling is evidence that AI has become easy to try, and the productivity case depends on something harder. Until the depth measures move, in our view, the 32% should not be quoted as progress on productivity.
Source: AI adoption survey 2026, including its executive summary and full report, researched by BMG Research and published on 29 October 2026. Fieldwork ran from 7 January to 21 April 2026, with 3,503 interviews among private sector businesses with five or more employees. The OECD figures come from Macroeconomic productivity gains from Artificial Intelligence in G7 economies (OECD AI Papers series, No. 41, June 2025), summarised by its authors in a SUERF policy brief. The Productivity Institute’s figures come from its release and the report’s summary site.
This strategic analysis was written by Resultsense, a UK-focused AI news and analysis publication. We will be watching whether the UK’s official measures of AI depth start to move alongside its headline adoption figures. Read more analysis at Insights, or get in touch.