How should a UK small business start using AI?
Short answer
A UK small business should start with one costly, repetitive problem, not with a tool. Check that the data the problem needs is usable and lawful to use, try an off-the-shelf product before building anything, run a time-boxed pilot against a measured baseline, then roll out with simple rules on data and human review. As of November 2026, the UK help that exists is mainly free training, Knowledge Transfer Partnerships and local business support.
By Dr. Dulan Dias, co-founder and CTO of Tellme AI
Key facts
- In September 2026, 30% of UK businesses reported using at least one AI technology, including 29% of businesses with fewer than 10 employees.
- Among UK businesses with 10 or more employees, AI use rose from about 12% in September 2023 to 35% in June 2026, although the ONS describes adoption as relatively shallow.
- Among UK businesses using AI to improve operations, 63% reported no change in headcount, 1% an increase and 6% a decrease, according to the ONS in July 2026.
- Since 28 January 2026, every adult in the UK can take free AI foundations courses on the AI Skills Hub, part of a government and industry goal to train 10 million workers by 2030.
- Knowledge Transfer Partnerships pair a UK business with an academic partner and a graduate associate for 12 to 36 months; UKRI says a small or medium business might typically contribute £35,000 a year.
- As of November 2026, Innovate UK's Smart grants are closed to new applications.
- Help to Grow: Management is a 12-week, 50-hour course for leaders of SMEs with 5 to 249 employees; the government funds 90% and each participant pays £750.
- Help to Grow: Digital closed to new applications on 2 February 2023.
Definitions
- SME
- A small or medium-sized enterprise. UK statistics usually count businesses with fewer than 250 employees as SMEs, while individual schemes set their own limits, such as 5 to 249 employees for Help to Grow: Management.
- Shadow AI
- Staff using AI tools the business has not approved, often by pasting customer or company information into consumer chatbots. It is a common early risk because it starts before anyone has written a policy.
- Pilot
- A time-boxed trial of an AI tool on one task, with a baseline measured beforehand and success criteria agreed before it starts.
- Knowledge Transfer Partnership (KTP)
- An Innovate UK programme that pairs a UK business with a university or research organisation and a graduate associate, who works in the business on an innovation project for 12 to 36 months.
Start with a problem, not a tool
The best first AI project is a task your business already does often, by hand, at a cost you can measure, where a wrong answer is cheap to catch. Answering routine customer emails, drafting quotes from templates, summarising calls, extracting fields from invoices and searching internal documents all fit. Decisions about people, such as credit, hiring or pricing for individual customers, do not belong in a first project.
Check whether AI is the right tool at all. Chapter 2 of From Model to Product, Problem Framing and Requirements, recommends comparing an AI system with simpler options first: rules, search, human review, an existing tool or a change of process. A bad process can look like a gap only AI can fill when the real issue is routing, ownership or missing documentation. Rules, machine learning or an LLM explains the choice.
Tellme AI, where Dr. Dulan Dias is co-founder and CTO, shows the value of a narrow first problem. It began as an AI guide that answers visitors' questions about museum objects, and by 18 September 2026 it had answered 3,777 questions with 1.28 seconds to an answer. One clear job made success easy to measure.
A practical sequence for a first AI project
Work through these steps in order. Each one is cheap compared with fixing a rollout that skipped it.
- Pick one problem and measure it now: hours per week, error rate, response time or cost.
- Check the data: where it lives, whether it is accurate and whether you may use it for this purpose.
- Buy before you build: try AI features in software you already pay for, then specialist products.
- Run a time-boxed pilot, for example six to eight weeks, with a few willing staff and success criteria agreed in advance.
- Measure against the baseline, including the time spent checking and correcting AI output.
- Decide whether to roll out, adjust or stop. Stopping a pilot that did not pay is a good result.
- Govern the rollout with a one-page AI use policy and a named owner.
Is your data ready?
AI tools are only as good as the information they can reach. Before a pilot, answer four questions: where the data lives (shared drives, email, a CRM or paper), whether it is current and consistent, who is allowed to see it, and whether it includes personal data.
If it includes personal data, UK GDPR applies to the AI use as to anything else. Record why you are using the data, explain it in your privacy notice and check whether you need a data protection impact assessment; the ICO's guidance says most uses of AI with personal data will. UK AI regulation for startups explains the rules, which apply to established businesses too.
A tidy-up is often the most valuable result of a first project. One source of truth for prices, policies or product information helps staff as much as it helps any AI tool.
Should you buy, configure or build?
Most small businesses should buy or configure rather than build. The options run from cheapest and fastest to most expensive and slowest.
If you reach the last row, how much it costs to build an AI MVP shows how to budget it, and RAG vs fine-tuning covers the main technical choice for assistants that answer from your documents.
| Option | Good for | Watch out for |
|---|---|---|
| AI features in software you already use | Drafting, summarising and searching inside email, documents, accounting or CRM tools | Which data the feature can see, and whether the vendor uses it to train models |
| A specialist AI product | A defined task such as call transcription, invoice extraction or customer chat | Contract terms, data location, lock-in and per-seat or per-use pricing |
| A configured assistant over your documents | Answering questions from your policies, manuals or product data | Keeping documents current, and testing answers before customers see them |
| Custom development | A process that gives you an edge and that no product handles | Cost, maintenance and the need for someone to own it after launch |
What UK support exists in November 2026?
The UK support that exists for AI adoption in November 2026 is mostly training, partnerships and advice rather than grants. Much of the advice online points to schemes that have closed. This is what was offered when checked in November 2026:
| Support | What it offers | Status |
|---|---|---|
| AI Skills Hub | Free online AI foundations courses for every UK adult, many under 20 minutes | Open since 28 January 2026 |
| Knowledge Transfer Partnerships | An academic partner and graduate associate working in your business on an innovation project for 12 to 36 months, part-funded by Innovate UK | Open, with rolling deadlines |
| Help to Grow: Management | A 12-week leadership course for SMEs with 5 to 249 employees; £750 per participant, 90% government-funded | Running; a general management course, not an AI course |
| Business support helplines | Free advice from the Business Support Service in England, Find Business Support Scotland, Business Wales and Invest Northern Ireland | Open |
| Start Up Loans | Government-backed loans of £500 to £25,000 at 7.5% fixed, for businesses trading under 5 years, with up to 12 months of free mentoring | Open |
| Innovate UK Smart grants | Open grant competition for innovation projects | Closed to new applications |
| Help to Grow: Digital | Discounts on approved software | Closed since 2 February 2023 |
How do you measure an AI pilot honestly?
Measure an AI pilot in the same units you recorded at the baseline, and decide what success means before it starts. Time saved per task is the usual headline, but count the time staff spend checking and fixing AI output, or the saving will look larger than it is. If the numbers do not justify a rollout, stop or pick a different problem; how to evaluate an LLM application covers deeper testing for systems you build yourself.
- Time per task, before and after, including review time.
- Error rate: how often output needed correcting, and how serious the errors were.
- Outcomes: response times, customer satisfaction, quotes won or invoices processed.
- Cost: subscriptions, usage fees and staff time.
- Adoption: whether staff kept using the tool when nobody asked them to.
The risks, and simple rules that manage them
Most small-business AI risks are ordinary business risks in a new form, and a one-page policy covers them.
| Risk | Rule that manages it |
|---|---|
| Staff paste customer or confidential data into unapproved tools | List approved tools and what data may go into each; use business accounts, not personal ones |
| Confident but wrong answers reach customers | A person reviews customer-facing output until error rates are measured and low; see reducing LLM hallucinations |
| Personal data used without a lawful basis or notice | Record the purpose, update the privacy notice and check whether a DPIA is needed |
| Decisions about people made by AI alone | Keep a person responsible for decisions about credit, jobs, prices or access to services |
| A supplier changes price or terms, or shuts down | Keep your data exportable, and avoid processes that only one tool can run |
| Selling AI-driven services into the EU | Check whether the EU AI Act applies to your use before launch |
Frequently asked questions
Is it safe for staff to put customer data into ChatGPT or similar tools?
Only under rules you have set. Consumer accounts on AI chat tools may keep conversations and, depending on settings and terms, use them to improve models, while business plans usually offer stronger terms. Decide which tools are approved, what data may go into each, and whether your privacy notice and data protection assessment cover it. Until then, keep personal and confidential data out.
How much should a small business spend on its first AI project?
Start with what you already pay for: many business software subscriptions now include AI features, so a first pilot can cost little beyond staff time. Budget the pilot in hours as well as pounds, because training staff and checking output is the main cost. Commit to larger spending, such as custom development, only after a pilot has shown a measured saving.
Are there government grants for small businesses to adopt AI in the UK?
As of November 2026 the national schemes focus on skills and partnerships rather than grants for buying AI tools. Innovate UK's Smart grants are closed to new applications, and Help to Grow: Digital closed in 2023. What exists is free training on the AI Skills Hub, Knowledge Transfer Partnerships and free advice through each nation's business support service, which also knows about local schemes.
Does the EU AI Act affect a UK small business?
Only if your AI use reaches the EU. The Act covers businesses outside the EU when they place AI systems on the EU market or when their AI system's output is used in the EU. A UK firm using off-the-shelf AI tools for internal work is unlikely to be affected; one selling an AI product or running an AI chatbot for EU customers should check its duties, starting with transparency.
Should a small business hire an AI specialist?
Usually not as a first step. Training the staff who own the work, using AI features in existing software and buying specialist products covers most early needs. A specialist earns their cost when you build custom systems or run several AI tools that need integration and oversight. For one substantial project, a Knowledge Transfer Partnership or a contractor brings expertise without a permanent hire.
What should a small business AI policy include?
Keep it to one page: which AI tools are approved; what data may and may not go into them; which outputs need human review before they reach customers; who owns AI use and answers questions about it; how to report a mistake; and when the policy is reviewed. Add a line on telling customers when they are dealing with an AI system, and review the policy whenever you adopt a new tool.
Will AI reduce the number of staff a small business needs?
So far, rarely. In the ONS's July 2026 analysis, 63% of UK businesses using AI to improve operations reported no change in headcount, 1% an increase and 6% a decrease. For most small firms the practical question is which routine tasks staff can stop doing, and what they should do with the time instead.
Sources
- ONS: Business insights and impact on the UK economy, 8 October 2026
- ONS: Artificial intelligence in UK businesses, 2023 to 2026 (20 July 2026)
- GOV.UK: Free AI training for all (28 January 2026)
- UKRI: Knowledge Transfer Partnership
- UKRI: Innovate UK guidance for specific funds
- Small Business Charter: Help to Grow: Management FAQs
- GOV.UK: Final opportunity for businesses to access Help to Grow: Digital
- GOV.UK: Business support helplines
About the author
Dr. Dulan Dias is co-founder and Chief Technology Officer of Tellme AI in Belfast; founder of EchonLabs, grown from 2017 to more than 60 people before its acquisition in 2026; holder of a PhD in AI and data science from Queen’s University Belfast; and the author of From Model to Product.