Research report · AI adoption

AI Use Is Rising but Still Shallow in Small Firms, and Reported Gains Are Larger When Firms Integrate It into Their Workflows

Official surveys find about one small business in four uses AI, against 66% to 86% in owner surveys. In four countries, 1% to 4% of firms have built it in.

In Motion Automation research team · 8 November 2026

From above at dusk, a manager sits still reading at a desk in a small Tampa front office while people and traffic outside blur into light.

Abstract

Most firms are small, and since 2023 chat tools that draft text and answer questions have put artificial intelligence (AI) within reach of any of them. Counts of how many small firms use it disagree. In 2026, surveys of US owners run by banks, software companies and business groups reported 66% to 86%, while official surveys, which weight a sample of firms to stand for all firms, found about one in four. We compared official statistics from nine economies with owner surveys and public reports. In every official series with repeated readings, small-firm use rose: in the European Union from 6.4% of firms with 10 to 49 staff in 2023 to 17.0% in 2025, in the United Kingdom from 9.1% to 28.7% of firms with fewer than 10 staff between 2023 and 2026, and in Canada from 5.9% to 19.9% of firms with 1 to 4 staff between 2024 and 2026. Large firms added more percentage points, though small firms grew as fast in proportion. Most use is shallow: 7% of US employer firms that use AI had fully integrated it, and in four countries 1% to 4% of all firms had built AI into how the business runs. Firms that use AI more widely report gains more often, though the reports are self-assessed and show association, not cause. Among Japanese small firms in one survey, 23% of those using AI across the whole firm reported a transformational effect, against 2% of those using it for isolated tasks. Swedish public reports find that support for small firms often ends before use goes deep.

Introduction

Most businesses are small. About three US firms in four have fewer than 10 employees1, and among European Union firms with 10 or more people, about 83% have fewer than 502. Since 2023, artificial intelligence (AI), software that does tasks that usually need human judgment such as drafting text, answering questions or recognising speech, has become cheap enough for any of these firms to try. Much of the new use is generative AI, software that writes text or makes images from a typed request, usually through a chat tool such as ChatGPT, Gemini or Copilot. Among European firms with 10 to 49 people, the share using AI that analyses written text rose from 2.3% in 2023 to 9.9% in 2025, while the share using machine learning to analyse data moved from 1.8% to 3.8%3. The tools cost little: among US owners who graduated from one growth programme, 73% of those using AI spent under $100 a month on it, including 16% who used only free tools4.

An owner who wants to know whether the business is behind meets numbers that disagree. Surveys of US owners published in 2026 by banks, software companies and business groups put AI use at 66% to 86%456789. In the same year an official survey found that 24.3% of US employer firms with 1 to 4 employees had used AI in the previous two weeks10. An official survey, in this paper, is one run by a national statistics office or government agency on a sample of firms drawn to represent all firms, with answers weighted so that the results stand for all firms of each size. The owner surveys are run or paid for by a company or business group and answered by owners drawn from online panels, customer lists or the graduates of a programme. Both kinds of number are quoted as the share of small businesses using AI.

Whether a firm uses AI says little about how. An owner who asks a chat tool to draft one email and a firm that has built AI into how it quotes, books and bills customers both count as users. We call the sequence of steps a recurring job takes, such as quoting, booking or invoicing, a workflow. To integrate AI into a workflow means building AI into the steps a recurring job already takes, inside the software that job already runs in. Depth of use, here, is how far a firm has gone from one person using one tool on one task toward AI built into its workflows across the business.

Owners deciding where to spend a few hours and a few hundred dollars need to know which numbers to trust and whether depth changes results. So do the journalists, chambers of commerce, trade bodies and policy analysts who write about small firms. We ask five questions. How many small firms use AI by official measures, and is the share rising? Why do headline numbers disagree? How deep does small-firm use go? Do firms that integrate AI into their workflows report larger gains than firms that use a chat tool on one task? And what keeps use shallow?

Official surveys in Europe, North America and Asia show small-firm use rising fast, to between 17% and 29% of small firms in the latest readings of the five official series in Figure 1. Large firms added more percentage points over the same years, though small firms grew as fast in proportion. Most small-firm use is a chat or writing tool on one task, and 1% to 4% of all firms have built AI into how the business runs. In every source we found that splits reported gains by how widely or intensively a firm uses AI, the firms that go further report gains more often, though the reports are self-assessed and the firms that go further may differ in other ways. The paper makes these contributions:

  1. An account of why owner surveys report 2.4 to 3.5 times the official rate, from surveys that print their question and sample (Section 2).
  2. Official small-firm AI statistics from nine economies, side by side, each compared only with itself over time (Section 3, Table 1).
  3. Published measures of depth, each labelled with its base (Section 4), and the evidence for and against larger gains with deeper use (Section 5).
  4. The small-to-large gap stated both in percentage points and as a ratio (Section 6), Swedish public findings on why use stays shallow (Section 7), and what a small firm can take from them (Section 8).
Figure 1. Share of small and large firms using AI in five official surveys. European Union: Eurostat’s survey on information and communication technology use in enterprises, firms with 10 to 49 against 250 or more people, 2021 to 20253. United Kingdom: Office for National Statistics Business Insights and Conditions Survey (BICS), 0 to 9 against 250 or more employees, September 2023 to September 202611. Canada: Canadian Survey on Business Conditions (CSBC), 1 to 4 against 100 or more employees, second quarter of 2024 to 2026121314. Singapore: Infocomm Media Development Authority (IMDA), small and medium-sized enterprises (SMEs) against large local enterprises, 2023 to 202515. United States: Census Bureau Business Trends and Outlook Survey (BTOS), 1 to 4 against 250 or more employees, each fortnight from 17 November 2025 to 18 October 2026 under the current question10. Each survey asks its own question (Section 2), so the panels show change within a country, not levels across countries.

Why small business AI surveys disagree

Each official survey sets its own size bands. Eurostat, the statistics office of the European Union, surveys firms with 10 or more people in the 27 member states, so its small firms have 10 to 49 people and firms under 10 are not counted2. The UK’s Office for National Statistics (ONS) reports firms with 0 to 9 employees, Statistics Canada firms with 1 to 4, and the US Census Bureau employer firms with 1 to 4. Singapore reports small and medium-sized enterprises (SMEs) as one group, and Australia counts firms with 0 to 4 people as micro. Every figure and table in this paper names the band it uses.

The questions differ too. Eurostat asks whether the firm used at least one technology from a list: AI that analyses written text, speech recognition, text or speech generation, image recognition, machine learning for data analysis, AI that automates workflows or supports decisions, and autonomous robots or vehicles, with generation of pictures, video or audio added in 20252. The ONS asks businesses which AI technologies they currently use11. Statistics Canada asks whether the business used AI in producing goods or delivering services over the last 12 months14. The US Census Bureau runs the Business Trends and Outlook Survey (BTOS), a fortnightly official survey of employer businesses, and from September 2023 to October 2025 it asked whether the business had used AI “in producing goods or services” in the last two weeks. On 17 November 2025 the Census Bureau revised the wording to ask about AI used “in any business function”16, where a business function is an area of the firm’s work such as sales and marketing, accounting or customer service. The broader wording roughly doubled measured use, from about 10% of firms in late 2025 to 20% in the first months of 202617. It also lifted the largest firms most: by the Census Bureau’s own estimate, use at firms with 250 or more employees “surged to 31% from 13%”, while firms under 20 employees moved to “around 18%”1. Under the old wording, firms with 1 to 4 employees used AI about as often as the largest firms, 0.4 points apart from September to December 2023 and 1.4 points apart from July to October 2025, by our averages of the fortnightly estimates18. The two series measure different things, and we do not join them.

Figure 2. Share of small firms, or their owners, reporting AI use, by survey, with who was asked. Light bars: surveys run or paid for by a bank, software company or business group. Middle bar: the Federal Reserve Banks’ Small Business Credit Survey, a convenience sample weighted to match all employer firms. Dark bars: official surveys, weighted to stand for all firms in the size band. Sources are listed under the chart and in the references.45678919101120

Owner surveys ask a person rather than a firm, and their wording reaches further. The US Chamber of Commerce asks “Which of the following technologies do you use at work?”8. Intuit, which makes the QuickBooks accounting software, asks “How frequently, if at all, do you use any AI-enabled tool for your business?” and counts daily, weekly or monthly use6. Thryv asks “Are you currently using artificial intelligence (AI) in your business, including software you have been using that now uses AI?”9. U.S. Bank asks owners whether they use generative AI for their business7. The samples come from online surveys and research panels (SBE Council, Thryv and the US Chamber), paid respondents from an online panel and from Intuit’s own customers (Intuit), and graduates of Goldman Sachs’s 10,000 Small Businesses programme, 16.5% of whom replied4. None of these reports states that its answers are weighted to stand for all small firms in the country.

Census Bureau economists compared their own survey with an executive survey that reads much higher and concluded that “the concentration of AI use in specific industries and at large firms as well as distinguishing between firm and worker use are the dominant factors to help reconcile differences in estimated AI use”1. The reference period matters as well: asked about the last six months instead of the last two weeks, 27.7% of firms reported AI use, against 17% to 18%1. So does who answers. Owners and top executives report more use than finance or administrative staff, though “even if all respondents were top executives or owners, the firm-weighted use rate would have been at most 21%”1.

One federal survey shows both readings at once. The Federal Reserve Banks’ Small Business Credit Survey, fielded from September to November 2025 among employer firms with 1 to 499 employees and weighted to match all such firms21, asked whether “the business or any of its employees” were using AI for work, counting “features of software you already use”19. Of these firms, 46% said yes. Among the users, 49% were experimenting, 44% had partially integrated AI and 7% had fully integrated it, and 21% called AI very important to producing their core goods or services. Multiplied out, about 10% of all employer firms used AI they called very important to their core output, close to the 10.0% that the old BTOS question found for 22 September to 5 October 202518, and 3.2% had fully integrated it. Figure 2 sets the owner surveys beside the official ones.

How many small businesses use AI

Every official series we found with at least two readings shows small-firm use climbing (Figure 1, Table 1). In the European Union, 6.1% of firms with 10 to 49 people used AI in 2021, 6.4% in 2023, 11.2% in 2024 and 17.0% in 20253. Part of the last step may come from the wider 2025 technology list: 8.1% of small firms used AI to generate pictures, video or audio, a category first asked that year. In the United Kingdom, the share of firms with 0 to 9 employees using at least one AI technology rose from 9.1% in September 2023 to 14.4% a year later, 23.1% in September 2025 and 28.7% in September 202611. A separate UK government survey of firms with 5 or more staff found use among firms with 5 to 9 staff rising from 14% in early 2025 to 27% in early 2026, and among firms with 10 to 49 staff from 16% to 34%20.

Table 1. Official and government surveys of AI use by firm size. Small-firm and large-firm rates are the share of firms in that size band using AI in the latest wave, with the earliest wave in brackets where a series exists. Size bands are numbers of employees or people employed.
EconomySurveyPeriodWho was asked, and about whatSmall firmsLarge firms
European UnionEurostat ICT usage in enterprises2025 (2021)Firms with 10+ people, 27 member states; any AI technology from a list10 to 49: 17.0% (6.1%)250+: 55.0% (28.4%)
United KingdomONS Business Insights and Conditions SurveySep 2026 (Sep 2023)Businesses of all sizes, most sectors; AI technologies currently used0 to 9: 28.7% (9.1%)250+: 50.9% (18.7%)
United KingdomUK government AI adoption researchJan to Apr 2026 (2025)3,503 private businesses with 5+ staff; using or piloting AI5 to 9: 27% (14%)250+: 52% (36%)
CanadaCanadian Survey on Business ConditionsQ2 2026 (Q2 2024)Businesses with employees; AI to produce goods or deliver services, last 12 months1 to 4: 19.9% (5.9%)100+: 27.8% (5.8%)
United StatesCensus Bureau BTOS, current question5 to 18 Oct 2026 (17 to 30 Nov 2025)Employer businesses; AI in any business function, last two weeks1 to 4: 24.3% (17.0%)250+: 43.6% (30.5%)
SingaporeIMDA Digital Economy Report2025 (2023)Enterprises; SMEs against large local enterprisesSMEs: 23.4% (4.2%)Large: 70.4% (44.0%)
SingaporeMinistry of Manpower AI SurveyJan to Mar 20262,560 firms with 10+ staff; adoption includes planningUnder 25: 23.9%501+: 76.4%
South KoreaSurvey of Business Activities, analysed by KRIVET2024 (2017)Firms with 50+ workers; any stage of the businessUnder 100: 6.2% (0.7%)300+: 18.0% (3.5%)
JapanCommunications Usage Trend SurveyAug 2025 (Aug 2023)Firms with 100+ staff; internet of things or AI systems100 to 299: 21.8% (13.6%)2,000+: 62.3% (43.3%)
JapanSMRJ survey of AI useNov to Dec 20251,647 SMEs, most with 30 or fewer staffSMEs: 20.4%Not surveyed
AustraliaABS Characteristics of Australian Business2024 to 2025All businesses; AI ticked on a technology list0 to 4: 10.7%; 5 to 19: 12.3%200+: 34.7%
TaiwanSME White Paper surveyJun to Jul 20251,207 SMEs with 10 to 199 staff; includes planning10 to 199: 7.4%Not surveyed

Sources: Eurostat (2026)3; Office for National Statistics (2026)11; Department for Business, Innovation, Science and Trade (2026, Figure 2)20; Statistics Canada (2024, 2026)1214; U.S. Census Bureau (2026)10; Infocomm Media Development Authority (2026, p. 11)15; Soon (2026)22; Jeong and Geum (2026, Table 1)23; Ministry of Internal Affairs and Communications (2026)24; SMRJ (2026, p. 7)25; Australian Bureau of Statistics (2026)26; Small and Medium Enterprise and Startup Administration (2025, p. 82)27. KRIVET is the Korea Research Institute for Vocational Education and Training; SMRJ is Japan’s Organization for Small and Medium Enterprises and Regional Innovation; ABS is the Australian Bureau of Statistics.

In Canada, use among firms with 1 to 4 employees went from 5.9% in the second quarter of 2024 to 12.8% a year later and 19.9% in 2026; firms with 5 to 19 employees trailed them, at 6.3%, 9.4% and 14.9%121314. In Singapore, use among SMEs rose from 4.2% in 2023 to 14.5% in 2024 and 23.4% in 202515. Singapore’s Ministry of Manpower found 23.9% of firms with fewer than 25 employees adopting AI in early 2026, a count that includes firms only planning to22.

In the United States, under the current BTOS question, use among employer firms with 1 to 4 employees rose from 17.0% in the fortnight starting 17 November 2025 to 24.3% in the fortnight starting 5 October 202610. Averaged over six fortnights to smooth sampling noise, the share went from 17.7% (November 2025 to February 2026) to 23.3% (July to October 2026). In May 2026 the Census Bureau reported that use had grown among firms with 20 or more employees “but didn’t change significantly among firms with fewer than 20 employees” between December 2025 and May 202616; after May, the 1 to 4 class rose from 19.1% to 24.3%. We leave out the fortnight from 19 October to 1 November 2026, the latest published, because that round added new questions and the share of firms answering “do not know” to the AI question rose from 9.6% to 13.9%.

Four more economies publish official numbers, on narrower groups of firms or from a single survey. In South Korea, a national survey that covers only firms with 50 or more workers shows use among those with fewer than 100 workers rising from 0.7% in 2017 to 6.2% in 202423. Japan’s Communications Usage Trend Survey covers firms with 100 or more employees and asks about systems that collect and analyse data, combining the internet of things with AI; among firms with 100 to 299 employees the share rose from 13.6% in 2023 to 14.3% in 2024 and 21.8% in 202524. A single 2025 survey by Japan’s agency for small firms, SMRJ, found 20.4% of SMEs using AI25. In Australia, 10.7% of firms with 0 to 4 people and 12.3% of firms with 5 to 19 reported using AI in 2024 to 2025, when only 1% of all businesses had done so in 2021 to 202226. In Taiwan, one 2025 survey found 7.4% of SMEs with 10 to 199 employees using or planning to use AI27.

These rates are not comparable across countries. The size bands, the questions, the period asked about and the sectors covered all differ, and some counts include firms that are only planning or piloting. Each survey can be compared with itself over time.

Most small-firm use is a chat or writing tool on one task

Where surveys ask what small firms do with AI, the answer is mostly writing. Among US firms whose workers use generative AI, 85% use it for writing and editing, and information search comes a distant second at 50%1. Of US firms that use AI in a business function, 57% use it in three or fewer of the 15 functions the survey lists, and 24% in only one1. Among Goldman Sachs programme graduates who use AI, 86% rely on standalone chat tools4. In Japan, 82.6% of SMEs that have introduced AI use generative AI25, and in Taiwan, 68.9% of adopters rely on outside, ready-made tools27. In the European Union, small firms’ growth since 2023 came mostly from text and language technologies, while machine learning, process automation and image recognition barely moved3.

Figure 3. Shares of firms reporting deep AI use, by survey. Top: share of firms that use AI. Bottom: share of all firms. Each label names the base and the measure, and measures differ by survey. Comprehensive adopters are a group that Census Bureau economists identify statistically, using AI across many business functions. Sources are listed under the chart and in the references.1195284329202522

Few small firms change how they work to use AI. Among US employer firms with 1 to 4 employees that use AI, 65.5% made no changes at all to use it, 15.0% developed new workflows, 13.4% trained staff and 3.1% paid vendors or consultants to install or integrate it; at firms with 250 or more employees the shares were 48.5%, 21.1%, 31.4% and 10.3%30. Among EU firms with 10 to 49 people that use AI, 18% used it for three or more purposes in 2025, against 39% at firms with 250 or more, by our division of Eurostat’s shares3.

Every survey that asks about depth finds a small minority at the deep end (Figure 3). Among AI users, 7% of US employer firms had fully integrated AI19, 18% of Goldman Sachs programme graduates described AI as fully integrated in their core operations4, and 8% of owners in the SBE Council’s panel said AI was involved in more than half of their operations5. In a separate Goldman Sachs poll, 14% of owners said AI was “fully embedded in their core operations”31. Of UK firms using AI, 12% say they use it extensively28. Across all firms, the deep end is smaller still. A UK government survey counts 2% of businesses with 5 or more staff as intensive adopters, firms that use bespoke or embedded AI constantly and call it highly important20. In Singapore, 3.8% of firms with 10 or more employees had AI “actively integrated” within their processes22; in Japan, 3.6% of SMEs had introduced AI across the whole firm against 16.8% in some work25; and Census Bureau economists classed 1% of US firms as comprehensive adopters using AI across many business functions1.

One measure points the other way. An ONS article from June 2026 reports that 17% of AI-using firms with 0 to 9 employees use AI extensively, against 9% of those with 250 or more, though the same sentence gives 10% for firms with 10 or more employees, so the bases are not fully clear32. In a survey of SMEs selling on online platforms in 12 countries, three quarters of them in Japan, the OECD, an intergovernmental body of mostly high-income economies, found that 56.6% of AI users applied it only to isolated tasks and classed 76% as novices29.

Reported gains are larger where firms integrate AI into their workflows

The evidence on gains is mostly self-reported: firms say whether AI raised their sales, cut their costs or made staff more productive. A pattern in such reports is an association, two things that go together, and it does not show that one causes the other. Firms that were already growing may also be the ones that adopt AI and go further with it, which economists call selection.

The best-sampled evidence comes from the Census Bureau’s AI supplement to the BTOS, sent to its full sample of 1.2 million US businesses from November 2025 to February 2026. Firms that used AI were 3.2 percentage points more likely to report above-average current performance, 8.3 points more likely to expect above-average future performance and 6.4 points more likely to report rising sales, comparing firms of the same size class, industry, state and period1. Of the measures of how widely a firm uses AI, “the breadth of functional adoption”, the number of business functions using it, “correlates most strongly with performance and revenue-related metrics”1. The authors ask readers to keep “a descriptive rather than causal interpretation”1.

Two surveys split reported gains directly by depth (Figure 4). In the UK government survey, 54% of intensive adopters reported lower operating costs, against 24% of all AI users, and 81% reported higher employee productivity, against 50%20. Most users saw no cost change: 64% of AI users reported no effect on operating costs, and intensive adopters also reported cost increases more often, 20% against 9%. In the OECD survey’s Japanese sample of 887 SMEs, 2% of firms using AI for isolated tasks reported a transformational effect, 4% of those using it in one function, 11% of those using it across several functions and 23% of those using it enterprise-wide29. The OECD notes that these shares are “based on self-reported perceptions rather than measured outcomes”. Japan’s 2026 White Paper on Small and Medium Enterprises reports the same direction with a measured outcome: firms that had used AI since 2019 to raise sales through work they could not do, or could not do well, on their own showed median growth in value added of 23.0% from 2019 to 2024, against 17.9% for firms that had not33. That is a comparison of medians, with no adjustment for other differences between the two groups of firms.

Figure 4. Reported gains by how widely or intensively firms use AI. Left: Japanese SMEs selling on online platforms that report a transformational effect from AI, by how widely they use it, from the OECD D4SME survey, fielded late 2025 to early 2026, n = 887, not representative of all Japanese SMEs29. Right: UK businesses with 5 or more staff that use AI, reporting reduced operating costs and increased employee productivity, all users against intensive adopters, who use bespoke or embedded AI constantly and call it highly important (2% of all businesses)20. All effects are self-reported.

Shallow use is not worthless. Controlled experiments, in which workers are randomly given a tool or not so that differences can be credited to the tool, find real gains on bounded tasks. Among 453 college-educated professionals given writing tasks from their own occupations, a chat tool cut the time taken by 40% and raised output quality by 18%34. Among 5,172 customer-support workers, an AI assistant raised issues resolved per hour by 15% on average, most for the least experienced35. Among 758 management consultants, those with AI completed 12.2% more tasks and worked 25.1% faster on tasks within the tool’s abilities, but were 19% less likely to produce a correct solution on a task outside them36. In each case the gain came on the task the tool was placed inside.

Other evidence limits the claim. Among small US employer firms using AI, the Federal Reserve Bank of New York found that firms reporting AI as fully or partially integrated, or as important to production, were “equally likely to expect higher relative to lower employment” as firms only experimenting21. Selection is visible in several sources: in Intuit’s US sample, firms aiming for fast growth paid for dedicated AI tools at 38%, against 16% of firms aiming for stability6, and in 2018 data on 850,000 US firms, AI adoption was more common among startups with signs of high-growth ambition37. In Denmark, where chat tools spread widely in workplaces, surveys linked to administrative records rule out effects on earnings or hours worked larger than 2% two years after ChatGPT’s launch38. Studies that use statistical methods to separate cause from selection have found that AI use, and its intensity, went with higher productivity in German firms39, but those data predate generative AI. No study we found isolates the effect of depth of use on outcomes at small firms.

Large firms added more percentage points, while small firms grew as fast in proportion

A gap between small and large firms can be stated two ways: the difference between their shares, in percentage points, or how many times larger one share is than the other, the ratio. The two can move in opposite directions, and both are needed here. In the European Union, the gap between firms with 250 or more people and firms with 10 to 49 grew from 22.3 points in 2021 to 24.1 in 2023, 30.0 in 2024 and 38.0 in 2025, and it widened between 2023 and 2025 in 25 of the 27 member states3. Over the same years the ratio fell from 4.6 to 3.2, because small firms started from a low base and their share nearly tripled. Denmark, where 74.5% of large firms used AI in 2025, is the main exception: its gap narrowed from 39.1 points in 2023 to 37.1 in 2025. Across the OECD countries that report the data, the gap reached 34.6 points in 2025, up from 23.4 in 2023, though the set of countries reporting changes from year to year29.

The same pattern holds elsewhere over the full length of each series. In the United Kingdom the gap between firms with 250 or more and 0 to 9 employees grew from 9.6 points in September 2023 to 22.2 in September 2026, while the ratio fell from 2.05 to 1.7711. In Canada the gap between firms with 100 or more and 1 to 4 employees went from none in 2024 to 7.9 points in 20261214. In South Korea the gap between firms with 300 or more workers and those with fewer than 100 grew from 2.8 to 11.8 points between 2017 and 2024, while the ratio fell from 4.8 to 2.9; the authors call that fall in the ratio closer to an illusion, because the small-firm share started near zero23. In Australia, large-firm use rose from 9% in 2021 to 2022 to 34.7% in 2024 to 202526. In the United States, averaged over six fortnights, the gap between firms with 250 or more and 1 to 4 employees grew from 12.9 points to 17.3, with the ratio steady at about 1.710.

The most recent year runs the other way in two places. In Singapore the gap between SMEs and large local enterprises narrowed from 48.0 points in 2024 to 47.0 in 2025, after widening from 39.8 in 202315. In Japan, among firms with 100 or more employees, the gap between the largest firms and those with 100 to 299 narrowed from 46.0 points in 2024 to 40.5 in 2025, after widening from 29.7 in 202324. The OECD expects the size gap in AI to persist, writing that “AI adoption will likely remain skewed towards large firms without increased support for adoption”40; its projection to 2030 extends a trend fitted to six yearly points for large firms, which the OECD itself calls an informed estimate rather than a precise prediction.

Swedish public reports find that support for small firms often ends before use goes deep

Sweden’s Agency for Economic and Regional Growth, Tillväxtverket, commissioned three reports between 2025 and 2026 on how small and medium-sized firms take up AI and what public and private support does for them. One maps the support available and studies six firms, with an emphasis on firms of up to 49 employees41; one interviews service firms, including six cases of firms at the leading edge42; and one interviews 11 support organisations and reviews earlier reports43. In Eurostat’s figures, Swedish firms with 10 to 49 people went from 22.1% using AI in 2024 to 30.7% in 2025, while firms with 250 or more went from 56.3% to 71.9%3. The reports explain why so much of that use stays shallow. We quote them in Swedish, followed by our translation.

The first report’s summary of its interviews is direct: “Komplexiteten i verksamheterna gör dock att AI-adoptionen i många fall slutar på ytlig nivå” (the complexity of the businesses means that AI adoption in many cases stops at a shallow level)41. Support arrives in short pieces: “I flera fall utgörs erbjudanden av punktinsatser i form av enstaka föreläsningar eller korta kurser” (in several cases the offers are one-off efforts in the form of single lectures or short courses)41, and “många insatser i dag är projektfinansierade och tillfälliga” (many measures today are project-funded and temporary)41. The help often stops too soon: “De studerade SMF som fått stöd för AI-initiativ ser i flera fall att konsultstöd och resurser tagit slut innan de sett positiva effekter av den satsning de fått stöd för” (several of the SMEs studied that received support for AI initiatives found that consultant support and resources ran out before they saw positive effects of the effort that was supported)41. A car-parts supplier of about 20 people could not find outside experts who understood a small firm’s needs and, even with support, could not afford consultants for the time the work needed; it put text automation into use, while its demand forecasting never became reliable enough to use41.

The reports place part of the problem with the consultants and part with the buyers. “Konsultmarknaden saknar förmåga att möta SMF i att utveckla och implementera rätt lösningar på företagsspecifika problem” (the consultancy market lacks the ability to help SMEs develop and implement the right solutions to firm-specific problems), partly because of “en bristande förståelse för mindre bolags verksamhetsbehov och förutsättningar” (a poor understanding of smaller companies’ business needs and circumstances)41. On the buyers’ side, “Mest framträdande är att SMF inte vet vad de behöver, vad de köpa eller hur” (most prominent is that SMEs do not know what they need, what to buy or how)43. Firms ask for help built around their own work: the first report recommends “modulära program med praktiska utbildningar där företag får arbeta utifrån den egna verksamheten tillsammans med AI-experter” (modular programmes of practical training in which firms work from their own business together with AI experts)41. The third report finds where benefit comes from: “För många SMF uppstår stora effekter när befintliga verktyg och tjänster används mer systematiskt, integreras bättre och kopplas tydligare till affärsmål och arbetssätt” (for many SMEs, large effects arise when existing tools and services are used more systematically, integrated better and tied more clearly to business goals and ways of working)43. It also stresses that firms should build basic skills in-house over time, to avoid long-term dependence on outside providers43.

These findings come from interviews and a handful of chosen cases, and the third report interviewed support organisations rather than firms. Surveys elsewhere point the same way. Among Japanese firms that had not taken up AI, 63.4% said they could not picture which work it would apply to33, and 83.3% of Japanese SMEs in another survey said they lacked information on examples of successful use25. In Taiwan, 63.9% of SMEs gave “no clear need yet” as their main obstacle and 88.1% had no AI strategy27. In the UK, firms already using AI most often named limited skills (67%) and limited knowledge of how AI could help (64%) as barriers20.

What a small firm can do with this evidence

The evidence above does not prove that integrating AI causes larger gains, and it cannot say what any one firm will get. It does suggest where an owner’s next hours are likeliest to pay back, and how to find out cheaply. Each step below follows from a finding in this paper.

  1. Pick one recurring job that already costs hours or money. The gains that rise with depth are tied to how many of a firm’s functions and workflows use AI (Section 5), and the Swedish reports find effects where tools are tied to business goals and ways of working. A job that repeats every week, such as answering quote requests, scheduling or chasing invoices, gives the step many chances to pay.
  2. Measure the job before changing it. Count the hours, the response time, the errors or the money lost over a few normal weeks. Most UK firms using AI reported no effect on costs (64%), and a firm can only tell which group it is in with a number from before.
  3. Build the AI step into the software the job already runs in. Most small-firm use sits in a chat window beside the work (Section 4). In the controlled studies, gains came on the task the tool was placed inside.
  4. Measure again for a fixed period set in advance, and keep or stop on the number. Self-reported gains are what the surveys hold; a firm’s own before-and-after count is better evidence for that firm.
  5. Budget for someone to keep it running after setup. Several of the Swedish firms studied lost support before they saw results, and the third report advises building basic skills in-house so the firm does not depend on outsiders for good.

Our team at In Motion Automation builds this kind of work for small firms.

Discussion

The official record is consistent on direction. Small-firm AI use has risen in every official series, several-fold within two to three years in the European Union, the United Kingdom, Canada and Singapore, mostly through chat and writing tools, and a few percent of all firms have built AI into how they operate. Large firms have added more percentage points, but small firms have grown at least as fast in proportion, which is the pattern expected when both groups follow the same path of adoption and large firms are further along it. The high owner-survey numbers answer a different question: whether anyone at a firm, often a growth-minded one, uses an AI tool for work at all, including features inside software the firm already has.

Why would gains rise with depth? One explanation is arithmetic. A chat window saves minutes on one person’s task each time someone remembers to open it, while a step built into a workflow runs every time the job runs, for whoever does it. The Census Bureau finding that breadth across business functions is the measure most closely tied to performance fits this explanation, as does the Swedish interview finding that effects arise when tools are integrated and tied to ways of working. Selection fits the same data: firms with the ambition, staff and systems to integrate AI may be the firms that would have done well anyway, and in the New York Fed’s data, firms that had integrated AI were no more likely to plan hiring than firms only experimenting. The controlled studies show that the tools do save time on bounded tasks, so shallow use has value; whether that value grows when the tool is built into a whole workflow at a small firm has not yet been measured.

The evidence has clear limits. Size bands, questions, reference periods and sectors differ across surveys, so we compare each series only with itself. Eurostat’s figures leave out firms with fewer than 10 people, and its 2025 wave added a technology to the list. The US series under the current question covers one year, and the change of question in November 2025 broke the earlier series. Japan’s official series covers only firms with 100 or more employees and combines AI with the internet of things, and Korea’s covers only firms with 50 or more workers. Depth is measured differently in every source, from “fully integrated” to “extensively” to statistically identified groups, and the one ONS breakdown by size runs against the rest. Nearly all evidence on gains is self-reported, cross-sectional and associational; the OECD’s Japanese sample is drawn from platform sellers and is not representative, and the Swedish findings rest on interviews and chosen cases.

Several tests could settle the open question. Official surveys that already ask about depth, such as the Federal Reserve’s integration question or the UK government’s intensive-adopter measure, could publish their results by size band and repeat them, so that small firms’ depth can be followed over time. Linking those answers to recorded sales and costs in later years, as the Danish and German studies did for use, would show whether deeper use at small firms is followed by measured gains. A small firm can run the same test on one job with a before-and-after count.

Methods and sources

We searched for official statistics that publish business AI use by firm size, from national statistics offices and government agencies, and kept every series with a small-firm band; series with at least two waves appear in Figure 1 and Section 3. We added government-commissioned surveys (the UK government’s AI adoption research, Japan’s SMRJ survey, Taiwan’s SME White Paper survey), the Federal Reserve Banks’ Small Business Credit Survey, and owner surveys from banks, software companies and business groups that publish their question or their sample. China has no official survey of AI use by firm size that we could find, and it is not included. Every source was read on or before 8 November 2026, in the original language; translations from Swedish, Japanese, Korean and Chinese are ours.

Rates are as published, with these exceptions, each marked in the text. EU gaps and ratios, the EU share of AI users using three or more purposes, and the UK, Canadian and Singaporean gaps are our arithmetic on published cells. The US six-fortnight averages are means of the published fortnightly estimates for firms with 1 to 4 and 250 or more employees: the first six fortnights of the current question (collected 17 November 2025 to 8 February 2026) and the last six before the excluded fortnight (collected 27 July to 18 October 2026). The old-question size gaps are means of the fortnightly estimates from the BTOS data service. The Federal Reserve all-firm shares multiply the share of firms using AI by the share of users in each depth group. Figures were drawn with matplotlib from the published values; the code and data files are kept with this paper.

Sources

  1. Kathryn Bonney, Cory Breaux, Emin Dinlersoz, Lucia Foster, John Haltiwanger and Aditya Pande, "The microstructure of AI diffusion: Evidence from firms, business functions, and worker tasks," U.S. Census Bureau, Center for Economic Studies Working Paper CES-WP-26-25R, 2026. www2.census.gov
  2. Eurostat, "Use of artificial intelligence in enterprises," Statistics Explained, 2025. ec.europa.eu
  3. Eurostat, "Artificial intelligence by size class of enterprise (isoc_eb_ai)," European Commission, 2026. ec.europa.eu
  4. Goldman Sachs 10,000 Small Businesses, "The AI economy is a small business economy," Goldman Sachs, 2026. goldmansachs.com
  5. Small Business & Entrepreneurship Council, "Small business technology use survey," SBE Council and TechnoMetrica, 2026. sbecouncil.org
  6. Intuit QuickBooks, "2026 AI impact report: Mapping adoption, use, and impact across small to midsize businesses in the US, Canada, the UK, and Australia," Intuit, 2026. digitalasset.intuit.com
  7. U.S. Bank, "The small business perspective: Growing through the pressure," 2026. usbank.com
  8. U.S. Chamber of Commerce Technology Engagement Center, "Empowering small business: The impact of technology on U.S. small business," U.S. Chamber of Commerce, 2026. uschamber.com
  9. Thryv, "AI and small business adoption survey 2026," 2026. assets.thryv.com
  10. U.S. Census Bureau, "Business Trends and Outlook Survey: Employment size class estimates," 2026. census.gov
  11. Office for National Statistics, "Business insights and impact on the UK economy: 8 October 2026," 2026. ons.gov.uk
  12. Statistics Canada, "Use of artificial intelligence by businesses and organizations in producing goods or delivering services over the last 12 months, second quarter of 2024 (Table 33-10-0825-01)," 2024. www150.statcan.gc.ca
  13. Statistics Canada, "Use of artificial intelligence by businesses and organizations in producing goods or delivering services over the last 12 months, second quarter of 2025 (Table 33-10-1004-01)," 2025. www150.statcan.gc.ca
  14. Statistics Canada, "Use of artificial intelligence by businesses and organizations in producing goods or delivering services over the last 12 months, second quarter of 2026 (Table 33-10-1167-01)," 2026. www150.statcan.gc.ca
  15. Infocomm Media Development Authority, "Singapore digital economy report 2026," 2026. imda.gov.sg
  16. Alexa Grundy, Cory Breaux and Bishal Khatiwoda, "Large firms with at least 20 employees biggest AI users," U.S. Census Bureau, America Counts, 2026. census.gov
  17. Erick Garcia Luna, "AI adoption in business grows steadily but unevenly," Federal Reserve Bank of Minneapolis, 2026. minneapolisfed.org
  18. U.S. Census Bureau, "Business Trends and Outlook Survey API: Collection periods 31 to 84," 2025. census.gov
  19. Federal Reserve Banks, "2026 report on employer firms: Findings from the 2025 Small Business Credit Survey," 2026. doi.org
  20. Department for Business, Innovation, Science and Trade, "Artificial intelligence adoption research 2026," UK Government, 2026. assets.publishing.service.gov.uk
  21. Will Aarons and Asani Sarkar, "AI adoption and employment expectations: Evidence from a survey of small business owners," Liberty Street Economics, 2026. doi.org
  22. Da Ren Soon, "Adoption of artificial intelligence among firms," Ministry of Manpower, Manpower Research and Statistics Department, 2026. stats.mom.gov.sg
  23. Ji-un Jeong and Ye-jin Geum, "Gieob-eun eodikkaji AI-reul sseugo inna [How far are firms using AI?]," The HRD Review, 2026
  24. Ministry of Internal Affairs and Communications, "Reiwa 7-nen tsūshin riyō dōkō chōsa hōkokusho (kigyō hen) [Communications usage trend survey 2025: Enterprises]," Ministry of Internal Affairs and Communications, Japan, 2026. soumu.go.jp
  25. Organization for Small & Medium Enterprises and Regional Innovation, Japan, "Chūshō kigyō no AI tō no rikatsuyō ni kakaru jittai chōsa [Survey of the use of AI and related technologies by small and medium enterprises]," SMRJ, 2026. smrj.go.jp
  26. Australian Bureau of Statistics, "Characteristics of Australian business, 2024-25 financial year," 2026. abs.gov.au
  27. Small and Medium Enterprise and Startup Administration, "2025 zhōngxiǎo qǐyè báipíshū [2025 white paper on small and medium enterprises in Taiwan]," Ministry of Economic Affairs, Taiwan, 2025. sme.gov.tw
  28. James Benford, "2026 Deane-Stone lecture, National Institute of Economic and Social Research, 4 November 2026," Office for National Statistics, 2026. ons.gov.uk
  29. Marco Bianchini, Marta Lasheras Sancho and Juan Ramon Larraín Aylwin, "Empowering SMEs in the age of AI: The 2026 OECD D4SME survey," OECD Publishing, OECD SME and Entrepreneurship Papers 78, 2026. doi.org
  30. U.S. Census Bureau, "Business Trends and Outlook Survey: AI supplement tables, 2026," 2026. census.gov
  31. Goldman Sachs, "Survey: Small businesses embrace AI, but need training and support to fully harness it," 2026. goldmansachs.com
  32. Office for National Statistics, "Artificial intelligence in UK businesses: 2023 to 2026," 2026. ons.gov.uk
  33. Small and Medium Enterprise Agency, "2026-nen ban chūshō kigyō hakusho: Gaiyō [2026 white paper on small and medium enterprises in Japan: Overview]," Small and Medium Enterprise Agency, Japan, 2026. chusho.meti.go.jp
  34. Shakked Noy and Whitney Zhang, "Experimental evidence on the productivity effects of generative artificial intelligence," Science, vol. 381, no. 6654, pp. 187 to 192, 2023. doi.org
  35. Erik Brynjolfsson, Danielle Li and Lindsey Raymond, "Generative AI at work," The Quarterly Journal of Economics, vol. 140, no. 2, pp. 889 to 942, 2025. doi.org
  36. Fabrizio Dell'Acqua, Edward McFowland, Ethan Mollick, Hila Lifshitz, Katherine C. Kellogg, Saran Rajendran, Lisa Krayer, François Candelon and Karim R. Lakhani, "Navigating the jagged technological frontier: Field experimental evidence of the effects of artificial intelligence on knowledge worker productivity and quality," Organization Science, vol. 37, no. 2, pp. 403 to 423, 2026. doi.org
  37. Kristina McElheran, J. Frank Li, Erik Brynjolfsson, Zachary Kroff, Emin Dinlersoz, Lucia Foster and Nikolas Zolas, "AI adoption in America: Who, what, and where," Journal of Economics & Management Strategy, vol. 33, no. 2, pp. 375 to 415, 2024. doi.org
  38. Anders Humlum and Emilie Vestergaard, "Still waters, rapid currents: Early labor market transformation under generative AI," National Bureau of Economic Research, NBER Working Paper 33777, 2025. doi.org
  39. Dirk Czarnitzki, Gastón P. Fernández and Christian Rammer, "Artificial intelligence and firm-level productivity," Journal of Economic Behavior & Organization, vol. 211, pp. 188 to 205, 2023. doi.org
  40. OECD, "OECD Digital Economy and Society Outlook 2026," OECD Publishing, 2026. doi.org
  41. PA Consulting Group, "Svenska små och medelstora företags AI-kompetens: En kartläggning av ekosystemet med fallstudier [Swedish small and medium-sized enterprises' AI competence: A map of the support system with case studies]," Tillväxtverket, Report 0508, 2025
  42. Magnus Burvall, "AI som förändringskraft i tjänsteföretag [AI as a force for change in service firms]," Tillväxtverket, Report 0513, 2025
  43. Maria Alsterskär and Frida Lindqvist, "Hur digitalisering kan stärka små och medelstora företags konkurrenskraft: En kunskapsöversikt med fokus på AI, data och cybersäkerhet [How digitalisation can strengthen the competitiveness of small and medium-sized enterprises: A review focused on AI, data and cybersecurity]," Tillväxtverket, Report 0583, 2026

Want us to look at your business this way?

We measure where calls, leads and paperwork slip through, in your own records, before we build anything.

Book a free call
ResearchDownload the PDF18 pages