The short version: New research from the Small Business Institute and Accio, Alibaba.com’s agentic AI platform, surveyed 1,000 UK small business owners in August 2026. It found 70 percent feel confident with generative AI while only 48 percent use it regularly, a 22-point spread the authors call an “AI productivity gap.” The headline breakdowns underneath that finding are shares of the 48 percent, not of all firms. Multiply them out and AI use in supply chain among UK small businesses is about 2.4 percent, not the 5 percent being quoted.
That matters beyond one study, because small business AI adoption gets reported with numbers anywhere from 17.7 percent to 87 percent depending on who is counting. Some of that spread comes from different definitions, and some of it is simply that these are different countries and different samples. Either way, a percentage quoted without its source and its exact wording attached is not a benchmark you can use.
What the study found
The research comes from the Small Business Institute, run by the UK campaign group Small Business Britain, together with Accio, Alibaba.com’s agentic AI platform for sourcing and supplier work. Fieldwork covered 1,000 UK small business owners in August 2026, and the findings were released on September 17. Of those owners, 70 percent are somewhat or very confident using generative AI and 48 percent use it regularly. Among the users, 74 percent apply it to content and marketing, 33 percent to data analysis, and 5 percent to supply chain.
On agentic AI, meaning tools that carry out multi-step work rather than answering one prompt at a time, 44 percent have heard of it and 5 percent use it regularly, while 68 percent want to learn more. Asked what they would want from it, 60 percent named improved work-life balance and 50 percent said freeing up time to win customers.
The 5 percent that is really 2.4 percent
The 74, 33 and 5 are shares of AI users, not of all small businesses. Nearly every writeup has repeated them as though they describe the whole population. Run the multiplication against the 48 percent who actually use AI regularly and the real penetration across all UK small firms looks like this:
- Content and marketing: 74 percent of users, about 36 percent of all small firms
- Data analysis and reporting: 33 percent of users, about 16 percent of all small firms
- Supply chain management: 5 percent of users, about 2.4 percent of all small firms
The direction of the study’s argument survives this correction. Marketing use really does dwarf operations use. But the gap is narrower than the raw percentages imply, and the smallest number in the study is roughly half as large as it appears.
Why the numbers run from 17.7 to 87 percent
Set this study beside the others published recently and the range is wide. It is also explainable, and the explanation sits in the wording and the sample rather than in anyone being wrong.
| Figure | Source | What it counted | Where | Sample |
|---|---|---|---|---|
| 87% | Bluehost and ListenLabs, May 2026 | Used at least one AI tool | US | 350 owners |
| 77% | Intuit 2026 AI Impact Report | Use AI regularly | US | 34,000 surveys plus 5.3m QuickBooks businesses |
| 61% | OnDeck and Ocrolus, Q2 2026 | Using AI | US | 805 working-capital borrowers |
| 48% | Small Business Institute and Accio, Aug 2026 | Use AI regularly | UK | 1,000 owners |
| 17.7% | JPMorganChase Institute, Dec 2025 | Ever paid for an AI service | US | 4.6m firms, payments data |
Two separate things are going on here. The distance between 87 percent and 17.7 percent is definitional. Bluehost counted anyone who has opened an AI tool once, while the JPMorganChase Institute counts firms that have actually paid for one, which excludes the very large number of businesses running free ChatGPT.
The distance between Intuit’s 77 percent and this study’s 48 percent is not definitional at all, because both asked about regular use. That one is a real difference of country and sample, and it is the plainer reminder of the two: a headline percentage does not travel across a border.
So the question to ask of any adoption figure is who was asked, where, and what counted as using AI. We have reported several of these ourselves, including Intuit’s 77 percent against the policy gap and OnDeck and Ocrolus at 61 percent alongside the margin data. They are all real. They are not interchangeable.
Who paid for the question
Accio co-produced this research, and Accio sells agentic AI for researching markets, evaluating suppliers and managing daily operations. The two gaps the study surfaces are that firms use AI for marketing rather than operations, and that agentic adoption is low while curiosity is high. Those are precisely the findings that make the case for the product, and the study’s lowest single number, supply chain use, is its sponsor’s core category. Alibaba.com Managing Director Michelle Lau frames current adoption as having been “at a surface level.”
None of that makes the data wrong. Vendor-commissioned research is a large share of what exists on small business technology, and refusing to read it would leave you with very little. It does mean reading the framing separately from the findings. The finding is that operations use is low. The framing is that the answer is an agent you can buy today.
The finding everyone buried
Coverage led with the line that owners want AI to win back time rather than just write copy. Underneath sits something more specific: asked what they wanted from agentic AI, more owners chose improved work-life balance, at 60 percent, than chose any commercial outcome. Not revenue, not growth, not cost. Time.
Rosemary Omeje, who founded Rosy Empire, said in the release that the tool took supplier research off her plate “while still allowing me to review the information and make the final decisions myself,” and that “for me, AI is about giving me back time.” She is a customer the vendor put forward, so treat it as illustration rather than evidence. But the pattern she describes, where the software does the legwork and the owner keeps the judgment, is the version of this that holds up in small businesses.
What to do with this
Place yourself on that table honestly. If you have an AI tab open but have never paid for anything, you are in Bluehost’s 87 percent and not in the JPMorganChase 17.7 percent, and the productivity gap this study describes is probably yours.
Cost is not the obstacle it once was. The JPMorganChase data puts typical entry spending at $20 to $30 a month, with median monthly spending among small business AI users falling from about $80 in 2022 to about $28 in 2025.
And the study’s core finding survives the correction. Almost everything small businesses do with AI is marketing and content, the most visible work and the easiest to check. The quieter operational jobs, reconciling suppliers, pulling numbers together, chasing information, are where the hours actually go, and they are worth a look before another content tool. For a structured way in, we laid out a 30-day path built on one task and one measurement.
Frequently Asked Questions
Which small business AI adoption number should I actually trust?
All of them, once you know what each counted and who was asked. Bluehost’s 87 percent counts US owners who have tried any AI tool. Intuit’s 77 percent counts US owners who use it regularly. This study’s 48 percent counts UK owners who use it regularly. The JPMorganChase Institute’s 17.7 percent counts US firms that have ever paid for AI. A figure quoted without that context is not a benchmark you can act on.
Does a UK study tell me anything if I run a US business?
The population figures do not transfer, and the JPMorganChase paid-adoption data is the better US benchmark. What does transfer is the shape: marketing use running far ahead of operations use, and confidence running ahead of regular use. Both show up in US surveys too.
What is agentic AI, in plain terms?
It is software that carries out a multi-step task rather than answering one question. Instead of asking a chatbot to draft a supplier email, you give it the goal of finding and comparing suppliers and it works through the steps. The useful version keeps you in the loop for the decision.
Should I be worried that only 5 percent use AI for supply chain?
No, and the real figure across all small firms is closer to 2.4 percent. Low adoption in a category tells you the tools are early or the fit is poor, not that you are behind. Watch the category; you do not have to buy into it this quarter.
Where do you land on that table, and if you have never paid for an AI tool, what has actually stopped you?
