At a glance

UK SME adoption of artificial intelligence is no longer marginal, but it is not yet deep or universal. The most recent ONS analysis, published in July 2026, reports AI use among around 35% of UK businesses with 10 or more employees, up from around 12% in late 2023. Other surveys report lower or higher figures because they ask different businesses about different kinds of use. Most adoption remains practical and limited, focused on administration, marketing and task automation. Accuracy, privacy, cost, skills and integration are the main obstacles.

What has happened

Artificial intelligence has moved from a specialist technology to something that many small businesses can access through ordinary software. Generative tools such as ChatGPT, Microsoft Copilot and Claude have made it easy for an employee to produce text, summarise a document, analyse a spreadsheet or create an image. At the same time, businesses have continued to use less visible forms of AI, including machine-learning data processing, image recognition, robotics and automated decision systems.

The most recent published national picture comes from the Office for National Statistics, whose July 2026 analysis of the Business Insights and Conditions Survey found that self-reported AI use among UK businesses with 10 or more employees had risen from around 12% in late 2023 to around 35%. That is a substantial increase, but the ONS described adoption as relatively shallow. The average number of AI technologies used by an adopting business rose only from around 1.4 to around 1.6 over the same period.

There is no single adoption figure that applies to every UK SME. The ONS measure covers businesses with at least 10 employees, so it does not provide a direct estimate for the many UK micro-businesses with fewer than 10 staff. Its survey also asks about a defined range of AI technologies. The Department for Science, Innovation and Technology’s 2025 research, based on 3,500 interviews conducted between February and May 2025, reported that one in six businesses currently used AI. That research asked the person responsible for technology and noted that it would not capture informal, unreported use by staff.

By contrast, a YouGov survey of 1,000 SME decision-makers in 2025 found that 31% of SMEs currently used AI and another 15% planned to do so. A separate YouGov study conducted from 24 February to 3 March 2026 found that 5% of SME decision-makers used AI extensively, while 29% used it in a more limited way. These figures are not necessarily contradictory. They use different samples, question wording and thresholds for what counts as use.

The safest conclusion is that a meaningful minority of SMEs are using AI, adoption has grown quickly since 2023, and extensive, organisation-wide implementation remains much less common than occasional or limited use.

Who this applies to

  • Micro-businesses with fewer than 10 employees: The ONS July 2026 estimate does not directly cover this group. A sole trader or very small firm may be using a general-purpose AI tool, but that is not the same as having a formal business-wide implementation.
  • Small businesses with 10 to 49 employees: These are included in the ONS measure and are likely to have more opportunity to benefit from shared processes, but may lack dedicated technical or data staff.
  • Medium-sized businesses with 50 to 249 employees: These are included in the ONS measure and are more likely than smaller firms to have the systems, data and management capacity needed to connect AI to established processes.
  • Businesses considering adoption: A plan to adopt AI is evidence of interest, not evidence of implementation. The ONS reported in March 2026 that 18% of businesses planned to adopt at least one AI technology within three months.
  • Businesses with informal staff use: An employee using ChatGPT for a draft or summary may not be recorded as the business using AI, particularly where there is no policy or central oversight.

Adoption is rising, but implementation is still shallow

The direction of travel is clear. The ONS’s July 2026 analysis found that reported use among businesses with 10 or more employees rose by roughly 23 percentage points between late 2023 and 2026, from around 12% to around 35%. The precise timing of the increase matters. Generative AI became widely visible after 2022, and much of the subsequent growth appears to have begun with easily accessible tools rather than large technology projects.

A separate ONS release, based on the late December 2025 wave of BICS, reported that 25% of businesses were using some form of AI, up 15 percentage points since the question was first introduced in late September 2023. In the late March 2026 wave, the ONS reported 26% using AI, up eight percentage points from late March 2025, while 18% planned to adopt AI in the next three months.

The difference between the 26% BICS figure and the ONS article’s figure of around 35% is important. The ONS article brings together data across the period from 2023 to 2026, while individual BICS releases refer to particular survey waves and may use different analytical populations, response patterns or measures. The ONS also cautions that BICS statistics are in development and that its questions can change. It is therefore better to treat these as evidence of a rising range of adoption, rather than as competing claims about one exact percentage.

Business size is another consistent dividing line. In December 2025, the ONS reported AI use of 44% among businesses with 250 or more employees. The ONS’s July 2026 analysis also found that larger firms were more likely to have adopted AI. The available sources do not give a comparable national percentage for micro-businesses, so it would be misleading to assume that the 35% figure describes them.

There are signs that adoption is moving beyond curiosity, but not yet towards deep integration for most firms. The modest rise in the average number of AI technologies used by adopting businesses, from around 1.4 to around 1.6, suggests that many businesses are adding one tool or one use case rather than redesigning several connected processes.

What a survey may count Typical meaning
Generative AI use Producing or editing text, images or other content with a large language model or similar tool
AI technology use A wider category that may include machine learning, image processing, robotics and autonomous systems
Planned adoption An intention to adopt within a stated period, not current operational use
Extensive use A higher threshold of use reported by a decision-maker
Informal use Individual staff activity that may not be recorded as a business system

What SMEs are using AI for

The practical uses reported by SMEs are concentrated in activities where a business already creates, sorts or responds to information. In the 2025 YouGov survey, 54% of SMEs using or planning to use AI said they were applying it to automate tasks. Marketing or advertising was reported by 45%, product or service development by 37%, customer service by 31%, and operations or logistics by 28%. Decision-making was reported by 19%.

For a small professional services firm, that may mean producing a first draft of a proposal, turning meeting notes into actions, or summarising a long document before a person checks it. For a manufacturer or engineering business, it could mean searching technical documents, classifying enquiries, supporting quality checks or reporting on production information. For a trade business, the immediate value may be less glamorous but more useful, such as following up an unanswered quote, extracting details from an enquiry or reminding a customer about an invoice.

Marketing is one of the most accessible entry points. AI can help create alternative versions of an advert, organise ideas for a campaign, draft website copy or turn a completed job into a case-study outline. The owner still needs to check whether the result is accurate, suitable for the brand and based on information the business is entitled to use.

Customer service and sales uses tend to involve speed and consistency. A system may classify incoming enquiries, suggest a response, identify the service requested or route a question to the right person. It should not be treated as an authority simply because it responds quickly. Pricing, technical promises and commitments to customers still require a controlled process.

Administration is another natural target. AI can assist with document processing, data entry, invoice information, reminders, scheduling and internal reporting. These uses are particularly relevant to firms where office staff spend time moving information between email, spreadsheets, accounting software and job-management systems.

The ONS found in its July 2026 analysis that improving business operations was the most common use of AI, reported by more than 60% of larger businesses. This is a useful reminder that AI adoption is not limited to chatbots. It includes changes to how information is handled and how repetitive work is organised.

The strongest early cases usually share three features: a repetitive task, information in a usable format and a clear human check. The weakest cases begin with a tool and search for a problem afterwards.

Industries are moving at different speeds

AI use varies sharply by sector. The ONS reported in July 2026 that 58% of businesses in information and communication used AI, compared with 13% in construction. In Scotland, the December 2025 BICS estimates reported by the Scottish Government showed a similar pattern, with 60.9% of information and communication businesses using some AI, compared with 28.9% among small and medium-sized businesses with 10 to 249 employees overall.

The YouGov 2025 SME survey also placed IT and telecommunications at the front, with 56% reporting AI use, followed by media, marketing and advertising at 53%. It reported lower adoption in real estate at 11%, transportation and distribution at 15%, hospitality and leisure at 18%, manufacturing at 19%, and retail at 19%.

These differences are plausible without implying that one sector is inherently more innovative than another. Information-intensive firms often have employees who work digitally, produce large volumes of text or code, and can test a tool without changing a physical process. A construction business may see clear potential in estimating, scheduling or document handling, but it also has to deal with site work, safety, drawings, subcontractors and fragmented systems. Manufacturing may need reliable integration with machinery and production controls before an experiment becomes operational.

The sector figures should still be read carefully. The ONS figures cover businesses with 10 or more employees, while the YouGov figures cover SME decision-makers and use a different survey design. They establish the broad pattern, not a league table with perfectly comparable results.

Why businesses are adopting AI

The commercial case is usually about capacity rather than replacing an entire role. A small firm may want to answer enquiries sooner, issue quotes more consistently, reduce time spent on routine administration or understand its sales pipeline without asking someone to build a report manually.

The YouGov study published in 2026 found that 51% of SME decision-makers wanted efficiency or productivity improvements to motivate a switch from traditional software to AI tools. Cost savings were cited by 45%, ease of use by 34%, better features and capabilities by 30%, and better integration with other tools by 28%.

That evidence describes motivations and preferences, not guaranteed results. AI may reduce the time needed for a task, but the business still has to specify the task, provide suitable information, check outputs and maintain the process. Poorly designed automation can move work rather than remove it, creating a new checking burden or sending incorrect information more quickly.

For firms with five to 50 staff, the most credible opportunity may be to increase the useful capacity of the existing team. If an office worker can spend less time copying enquiry details, chasing approvals or preparing routine reports, the business may handle more work without immediately adding headcount. That is an operational choice, not proof that AI has raised productivity across the company.

The ONS’s employment findings support a cautious view. In late March 2026, 5% of businesses using AI reported a reduction in workforce headcount because of AI, rising to 7% among businesses with 10 or more employees. In late December 2025, the comparable figures were 4% and 5%. These figures do not show widespread job replacement. They also do not prove that no roles will change. They suggest that, so far, businesses more often describe AI as a way to alter tasks than as a direct reason to remove whole workforces.

The barriers between a trial and a working system

The first barrier is knowing what to do with the technology. DSIT’s 2025 research found that one in six businesses used AI, while most had no active plans to adopt it. Its interviews explored the difference between recognising AI’s potential and finding a specific business problem worth solving.

Accuracy is a particularly practical concern. In the 2026 YouGov research, 65% of SME decision-makers who used software identified reliability and accuracy as a barrier to adopting AI. Data security and privacy were cited by 51%, regulatory and compliance issues by 33%, the cost of switching by 30%, and training requirements by 29%.

The concerns are not limited to firms that have done nothing. The 2025 YouGov survey found that 49% of businesses not planning to use AI were concerned about data privacy and security, while 30% did not see the value. Among adopters, 42% worried about legal risks, 34% about job losses, and 58% about AI reducing business creativity.

Data maturity is part of the problem. The UK Business Data Survey 2026, covering businesses surveyed between October 2025 and January 2026, found that 41% of businesses handling digitised data reported using AI for at least one purpose. It also found that 17% of AI-using businesses had no AI policy, while formal written policies were more common among large businesses, at 56%.

For a small business, the implementation questions are often more important than the model itself. Where is the customer information stored? Can the system connect to the existing accounts, website, pipeline or job-management software? Who checks an output before it reaches a customer? What happens when a supplier changes a format or an employee leaves?

The answer will differ by size. A very small firm may be held back by cost, time and the absence of anyone who owns the project. A larger SME may have more resources but face integration, permissions, governance and staff training problems across several departments.

The practical distinction is between shadow AI and managed adoption. Shadow AI is when employees use public tools without a company policy, approved account or management oversight. It can be useful evidence that staff have found a real need, but it can also expose confidential information and create inconsistent results. Individual experimentation is not the same as connecting AI to the process that produces quotes, invoices, reports or customer replies.

What government support can and cannot solve

The government’s approach is to encourage wider use through infrastructure, data, skills and business adoption. DSIT’s 2025 research was commissioned to improve the evidence for that policy, following the AI Opportunities Action Plan published in January 2025. The research also refers to a free AI training initiative for small businesses launched in January 2026.

For an SME, the practical value of such support depends on whether it helps staff identify a suitable use, understand data responsibilities and implement it safely. Training that only demonstrates prompts may increase experimentation without improving the underlying workflow.

The regulatory position is also relevant, even where a business is not building an AI system. The 2026 UK Business Data Survey found that 53% of AI-using businesses were aware of regulatory guidance, but only 19% found it clear. Among businesses handling digitised personal data, 46% agreed that Information Commissioner’s Office guidance was clear and easy to understand.

That uncertainty is a reason to document how a tool is used, limit access to sensitive information and keep a human accountable for consequential decisions. It is not a reason to assume that every AI project requires a specialist compliance department. The right level of control should reflect the risk of the task.

Where this falls short

  • The ONS’s strongest comparable trend data covers businesses with 10 or more employees, so it does not directly measure micro-business adoption.
  • Surveys define AI differently. Some include machine learning and robotics, while others focus on generative tools or ask about extensive use.
  • Self-reported use does not show how often a tool is used, whether it is accurate, or whether it has produced a measurable financial benefit.
  • The available figures describe adoption and reported effects, not a controlled estimate of productivity gains.
  • Survey results can capture formal business use more reliably than informal employee activity, so shadow AI may be under-recorded.
  • Sector comparisons use different surveys and populations, and should be treated as directional rather than perfectly like-for-like.

Worked example

Illustrative example: a small engineering services firm with 12 employees receives enquiries by email, prepares quotes in a spreadsheet and sends invoices through separate accounting software.

In the first month, the firm could use an AI-assisted process to extract the customer name, requested service, deadline and key requirements from incoming enquiries. A member of staff would check the extracted information before it entered the sales pipeline. The same process could flag enquiries that had not received a response after a set period.

In month two, the firm could use its approved information to prepare a first draft of a quote summary and a follow-up message. The person responsible for the quote would still confirm scope, price, delivery timing and technical wording. The system would not be allowed to invent a specification or send a commitment without approval.

In month three, the firm could connect completed-job information to a simple dashboard showing open enquiries, quotes awaiting decisions and invoices needing attention. The value would not come from asking a chatbot general questions. It would come from reducing repeated copying and making the next action visible.

The firm should compare the process before and after implementation. It could record how long enquiry handling takes, how many quotes are overdue and how often staff correct extracted information. If the checking work is greater than the time saved, or if the data cannot be kept secure, the project should stop or be redesigned.

This example is deliberately modest. It does not assume a headcount reduction or a guaranteed productivity gain. It shows the type of connected, supervised use that is more meaningful than giving every employee an unstructured AI account.

Implevo's View

The evidence supports action where a business has a repeated administrative problem, reliable information and a clear owner for checking the result. AI is worth testing for tasks such as enquiry handling, quote follow-up, document processing, reporting and routine customer communications. It is not worth adopting simply because competitors are discussing it, or where the proposed system cannot be trusted with the data and cannot connect sensibly to existing work.

The next stage is likely to be less about replacing ordinary software and more about adding AI inside it. The 2026 YouGov study found that 54% of SME decision-makers did not believe AI would replace traditional software platforms within three years, while 80% of businesses using traditional software said they were satisfied with it. That points towards gradual integration, with AI helping existing systems handle parts of a process rather than removing them altogether.

Owners should start with one measurable bottleneck, establish what information the tool may use, and decide who approves its output. Implevo can help investigate that through a fixed-scope Discovery Day, if a practical conversation about the business problem would be useful.

References

  1. AI in UK businesses - GOV.UK, gov.uk, 20 July 2026
  2. Business insights and impact on the UK economy - Office for National Statistics, ons.gov.uk, 1 April 2026
  3. Artificial intelligence in UK businesses: 2023 to 2026, ons.gov.uk, 20 July 2026
  4. Business insights and impact on the UK economy - Office for National Statistics, ons.gov.uk, 8 January 2026
  5. AI Adoption Research, gov.uk, 13 February 2026
  6. UK Business Data Survey 2026 - GOV.UK, gov.uk, 18 June 2026
  7. Will AI replace traditional software tools for small businesses in the UK?, yougov.com, 5 May 2026
  8. Artificial Intelligence - Business Insights and Conditions in Scotland (wave 147): 15 January 2026 - gov.scot, gov.scot, 15 January 2026
  9. We polled UK SME leaders about AI adoption. Here's what ..., yougov.com, 7 August 2025