In short: Small-firm AI adoption has flattened while large firms pull away, but cost is cited by only 4% of non-adopters and skills by 3% - four in five say AI simply is not applicable to their business, which is the one barrier you can test cheaply.
You have read that AI adoption is booming. Ninety percent of this, eighty percent of that. Yet the official figures on small business AI adoption say it has flattened.
Meanwhile, here is what probably happened in your business. Someone tried ChatGPT. It wrote a decent email. Nobody could see how it applied to the scheduling, the supplier chasing, the invoicing - the things that eat the week. So it quietly stopped being used, and you concluded AI was for other kinds of companies.
That instinct is now visible in official statistics, and it is not a failure of nerve. But the reason small firms are falling behind is not the one you have been sold, and the data is unusually clear: it is not cost, and it is not skills.
Every figure below comes from a national statistics office, a central bank or the OECD, with sample sizes stated. This is one of the few AI topics where genuinely good data exists.
1. The stall is real, and here is its actual shape
The US Census Bureau’s Business Trends and Outlook Survey samples around 1.2 million businesses fortnightly. Its finding, published 26 May 2026 and covering December 2025 to May 2026, is blunt: AI use increased among firms with at least 20 employees but did not change significantly among firms with fewer than 20. Fewer than 20% of firms with four or fewer employees reported using AI, against 32% at 100 to 249 employees and 37% at 250 or more.
The same pattern holds across the Atlantic. Eurostat’s 2025 survey, covering around 157,000 enterprises, found AI use at 17.0% of small enterprises, 30.4% of medium and 55.0% of large. The OECD, pooling national statistics across member countries, puts it at 11.9% for firms with 10 to 49 employees against 40% for firms with 250 or more.
One caveat worth knowing, because it flatters the headline: Eurostat’s survey excludes firms with fewer than 10 employees entirely. Most of the world’s small businesses are not in that 20% figure, so the true micro-business number is almost certainly lower.
And this gap is unusual. The OECD makes a comparison nobody else does: small firms were about half as likely as large ones to buy cloud computing or use connected devices, but they are less than one-third as likely to use AI. AI is diffusing less evenly than the technologies before it.
The pattern repeats well beyond the US and Europe. Australia’s Bureau of Statistics, surveying nearly 7,000 businesses, puts firms under 20 staff at roughly 11% against 35% for firms of 200 or more. Singapore’s regulator reports 14.5% of SMEs against 62.5% of large enterprises. Japan shows the widest gap found anywhere: 6.8% against 34.6%, a factor of five.
The smallest firms are not the problem
Here is the part that overturns the usual framing. Statistics Canada asks this question quarterly with size bands, and its Q2 2026 data shows adoption is not a straight line. Firms with 1 to 4 employees sit at 19.9%, above firms with 5 to 19 employees at 14.9%, with 20 to 99 at 25.8% and 100-plus at 27.8%. Canadian government econometric work published in April 2026 found the same shape independently: a U-curve with its minimum at around three employees.
So the laggard is not the solo operator. It is the small team. A one-person business is the decision-maker and can start this afternoon; a twelve-person business has staff, process, liability and customers depending on it, and no IT function to lean on. If you sit in that band, the friction you feel is structural.
2. The depth gap is bigger than the adoption gap
Headline adoption understates the problem, because it counts a firm using one tool the same as a firm using five.
On that measure the distance is much larger. Among large EU enterprises, 44.0% use at least two AI technologies and 33.6% use at least three. Among small enterprises, 10.6% use at least two and 6.5% use three or more. On headline adoption large firms lead by roughly three times; on multi-technology use, by more than four.
Worse, the rate of change differs. Large EU enterprises added 14.8 percentage points on two-or-more technologies and 13.5 points on three-or-more in a single year. Large firms are compounding. Small firms are adding a tool.
The risk for a small business is not being late to AI. It is being lapped by competitors who are stacking capability while you are still evaluating one tool.
One genuine counterweight, from the UK’s Office for National Statistics in July 2026: among businesses that do use AI, 17% of those with under 10 employees reported extensive use, against 9% of those with 250 or more. Fewer small firms adopt, but the ones that do often use it harder than the giants. Size predicts whether you start, not how committed you are once you have.
Whether that gap is widening depends entirely on how you measure it. UK data shows it most starkly: on our own calculation from the size bands the Office for National Statistics publishes, the distance between firms of 250 or more employees and firms under 10 went from 9.6 percentage points in September 2023 to 20.4 in June 2026, more than doubling. As a ratio it fell, from 2.05 times to 1.72. EU data splits the same way. One dataset, two opposite headlines, both accurate, so anyone quoting one without the other is selling you something. ONS claims neither, and its AI figures are official statistics in development rather than accredited ones.
One gap is not ambiguous, though. Asked how much they had invested in adopting AI over the previous twelve months, 36.2% of UK firms with under 10 employees said nothing at all, against 8.6% of firms with 250 or more. And 70.3% of the smallest firms had no plans to adopt in the next three months, against 35.2% of the largest. Adoption rates may be converging in relative terms. Commitment is not, and commitment is what produces the compounding described above.
Why aren’t small businesses using AI?
Not because of price. Among US firms with no plans to use AI, the reasons given were: AI is not applicable to this business, 80.9%. Lack of knowledge of AI’s capabilities, 7.3%. Privacy or security concerns, 6.6%. Too expensive, 4.1%. Lack of skilled workforce, 2.9%. Lack of required data, 2.2%.
Read that list twice. Four in five said the technology does not apply to what they do.
That is a US figure from Census research, and the OECD found the same thing independently on other continents: the top-cited barrier among non-users in both Canada and the United Kingdom was that generative AI is not suited to the type of work the company does. It was not hostility either - 86% of SMEs reported a neutral or favorable attitude to it.
Statistics Canada’s own barrier data, from a stratified sample of 9,251 respondents in April and May 2026, lands in exactly the same place. Asked what was stopping them, 40.0% of Canadian businesses said AI is not relevant to what they produce - rising to 41.4% among firms with 1 to 4 employees and falling to 21.3% among firms of 100 or more. Cost was cited by 10.6% and cybersecurity or privacy by 13.4%. Two national statistics offices, two continents, same answer.
Britain gives a blunter version of it. When the ONS asked in June 2026 what was preventing or delaying adoption, the largest single answer among businesses with fewer than 10 employees was not cost, at 6.5%, nor level of expertise, at 7.0%. It was that they had not attempted to use AI at all: 23.2%. For the smallest firms the obstacle is not a wall they have run into. It is a blank where the attempt should be.
European data points the same way. Asked why they had not adopted, most reasons EU firms gave barely varied by company size, including expertise (70.9% of small firms against 65.1% of large). The one that genuinely rose as firms got smaller was “AI is not useful for our enterprise”: 19.2% against 9.8%.
Why this matters more than a cost barrier
A cost barrier is honest and hard. A relevance judgment is a hypothesis, and most owners formed it after trying a general chat tool on general tasks - which tells you almost nothing about whether AI applies to your own operations. It is the cheapest barrier in the list to test, and the only one most businesses have never tested.
3. What the small firms getting returns did differently
The finding that matters more than any adoption percentage comes from the OECD’s survey of SMEs across G7 countries.
Among small firms using generative AI, only 29% report using it in their core activities. The rest have it on peripheral work - marketing copy, email drafts, tidying documents. And the report is explicit about the consequence: SMEs using generative AI for tasks considered core to the company were generally more likely to report benefits.
So seven in ten small AI users have it parked where it cannot move the business, then conclude AI does not do much. Both halves are true at once.
US Census research from April 2026, on a nationally representative sample, corroborates this from the other side: commercial performance correlated with the breadth of AI integration across functions and tasks. Yet among adopters, 57% had AI in three or fewer business functions.
A real number from our own deployment
Before building anything for the D2C business in our 68% cost reduction case study, we ran a cost attribution analysis. It found 70% of the operational budget going to predictable, rule-based work, and roughly £140,000 a year of staff time across three functions that turned out to be about 85% rule-following and 15% genuine judgment. That business would have told you, sincerely, that AI did not apply to its operations. The 85% was invisible until somebody counted it.
The other consistent marker is training, and it is where small firms are weakest. Across every G7 country, fewer than 30% of SMEs using generative AI report that employees participate in any AI-related training - from 11.3% in Japan to 29.4% in Canada. Canada’s quarterly data makes the size split stark: 24.0% of firms with 1 to 4 employees had trained existing staff on AI, against 68.1% of firms with 100 or more. So large Canadian firms are only about 1.4 times more likely to use AI than micro firms, but nearly three times more likely to train anyone on it. The adoption gap is narrow; the capability gap is wide, and that is the one that compounds. The OECD is clear-eyed about why this is structurally harder when you are small: fewer staff means less slack to release someone from revenue-generating work, and unit training costs are higher.
Myth vs Facts
Myth: “Most small businesses already use AI - I saw 58%.”
Fact: That figure (US Chamber of Commerce, August 2025, n=3,870) defines small business as under 250 employees and asks about self-identified use. The Census figure of roughly 18% uses a probability sample and asks whether AI was used in producing goods or services in a specific two-week window. Different populations, different questions, different sampling. Neither refutes the other, and averaging them is meaningless.
Myth: “AI adopters are 15% more productive, so adopting AI will make us 15% more productive.”
Fact: The OECD reports adopter premia above 4% and sometimes above 15% - and then states plainly that the link is explained “to a considerable extent” by the fact that firms which are already more digital and more competitive are also more likely to adopt. Once you control for that, the advantage “relevantly shrinks”. This is the most misused statistic in the small-business AI genre.
The cleanest test comes from Statistics Canada and Innovation, Science and Economic Development Canada, published April 2026, using survey data linked to administrative business microdata rather than opinions. AI adopters showed 16.8% higher productivity. Controlling for productivity that already existed before adoption cut it to 10.2%. Adding complementary capabilities cut it to 5.1% and it was no longer statistically significant. Their conclusion: gains take time and depend more on the surrounding investments than on adopting AI itself.
Myth: “AI is too expensive for a business our size.”
Fact: Cost ranks fourth among the reasons US firms give for staying out, and in EU data large firms cite it almost as often as small ones (34.8% against 38.8%). Price is a real constraint on ambition, but it is not what is keeping small firms out.
Myth: “AI levels the playing field against bigger competitors.”
Fact: Small firms themselves do not report this. In OECD survey data, SMEs rank improved employee performance first, then cost savings and performing new tasks - while offering new products, competing with larger companies and increasing revenue were among the least reported benefits. The OECD’s own forward view is that divides between leading and other firms may widen.
The four “it doesn’t apply to us” beliefs, tested
Almost every relevance judgment we hear is one of these. None survives contact with a count of where the week actually goes.
| The belief | What the data or our experience says | Where to start |
|---|---|---|
| “Our work is too bespoke to automate” | Bespoke output often comes from repeatable process; we found 85% rule-following in work assumed to be judgment | Log one week of one role, then sort the tasks by whether a rule decides them |
| “We are too small to justify it” | Cost is cited by 4.1% of non-adopters; large firms report it almost equally | Price the smallest version of one task, not a platform |
| “We would need someone technical” | Skills cited by 2.9%; the fastest-diffusing AI is the natural-language kind requiring no IT function | Train existing staff on one workflow before hiring anyone |
| “We tried it and it did not help” | 71% of small AI users have it on peripheral tasks; core-task users report more benefit | Re-aim it at a core, repeating, expensive task |
| “AI is for customer-facing marketing” | Marketing is the one function where small firms already lead - so it is the least likely place to find advantage | Look at back office: invoicing, reconciliation, scheduling, supplier chasing |
What this means if you are running a small business
The honest version of the opportunity is narrower than the marketing version.
There is one place the evidence genuinely favors small firms, and it is not competing with incumbents. It is filling gaps you could not fill by hiring: 39% of SMEs that use generative AI and had recently experienced a skills gap said generative AI helped compensate for it. If you have a role you cannot recruit for, or a function that is one person deep and fragile, that is where the case is strongest.
One honesty note, because we would rather you heard it from us. Almost all the returns evidence is observational and, as the OECD says, confounded by who adopts. There is exactly one field experiment on real small-business owners, published in Management Science on 10 July 2026, and its result is sobering: across 640 owners given an AI business mentor, the authors could not reject a zero average effect on profits. Owners already performing well gained over 15%, while those performing poorly did nearly 10% worse. The AI’s advice was not the difference; which advice owners acted on was.
- Count where one week actually goes in your most expensive function, in hours and money
- Sort those tasks by one question: does a rule decide the outcome, or does judgment?
- Pick the single most repeated, most expensive rule-following task - not the most interesting one
- Aim at core work, not peripheral work; the peripheral version is why most small firms conclude AI does nothing
- Budget training time for the people who will use it, however small the firm
- Add a second capability only once the first is running unattended, because compounding is what large firms are doing
- Write down the number you expect to move before you start
If you are weighing what to deploy, the difference between AI skills, agents and an agentic operating layer is the next decision, and whether to build, buy or just use a chat tool depends on how often the work repeats. Both matter less than picking the right task.
Is AI genuinely irrelevant to your business, or untested?
Tick each one that is true today.
- We concluded AI did not apply to us after trying a general chat tool
- We have never counted how much of our week is rule-following work
- Our AI use, if any, is on marketing or admin rather than core operations
- Nobody in the business has had any AI training
- We use exactly one AI tool, and have used one for over a year
- One critical function depends on a single person we could not easily replace
- We could not name the cost line we would expect AI to reduce
If this were your business, here is our first move
We would not recommend a tool. We would spend a week counting - where the operational budget goes, which of that work is decided by a rule, and what the expensive parts of it cost you in a year.
(If the answer involves your website, the companion question is how exposed your traffic is to AI search.) That exercise turns “AI does not apply to us” from a belief into a tested answer, and sometimes the answer is genuinely no. We have told businesses that. It beats buying software to find out.
But the statistic to sit with is the 80.9%. Four in five firms outside AI have decided it is not for them, most without measuring, while larger competitors add a capability every few months. The gap is not being created by budgets. It is being created by an untested assumption. If you would like help testing yours, talk to us - and if the honest answer turns out to be that your work is already lean, we will say so.
