Everyone's using it. Not everyone's getting much from it.
In March 2026 the British Chambers of Commerce reported that 54% of the firms it surveyed were actively using AI, up from 35% the year before and 23% in 2023. That's a quick change. What the headline can't tell you is how much of that use is doing anything useful.
It's a familiar picture. A ChatGPT subscription or two, Copilot switched on because it came with Microsoft 365, someone in marketing using it for social posts, and no clear idea whether any of it has saved time or money. That isn't failure. It's what happens when a tool arrives before a plan.
Start with the job, not the tool
The useful question isn't 'what can AI do?'. It's 'what do we do every week that's repetitive, mostly text or data, and easy for a person to check?'. Those three conditions are where today's AI is reliably good.
Walk through a normal week and write down the tasks that fit. For a small firm, the list often looks like this:
- Drafting replies to common enquiries, for a person to edit and send
- Turning meeting notes or call recordings into actions and follow-up emails
- Writing first drafts of quotes, proposals and job adverts from a template
- Summarising long documents, tenders or contracts before someone reads them properly
- Tidying and categorising spreadsheets, CRM records or product data
- First drafts of product descriptions, article outlines and social posts
Keep it away from the expensive mistakes
Notice what isn't on that list: anything where a wrong answer is costly and hard to spot. AI tools still get things confidently wrong. Keep them away from final figures, legal or medical advice, and anything you'd be embarrassed to send without checking. The rule of thumb is simple. If a person can check the output faster than they could have produced it, AI is probably helping. If checking takes as long as doing, it probably isn't.
Use what you already pay for first
Before buying anything new, look at what's already in your subscriptions. Microsoft 365 and Google Workspace now include AI features in many of their plans, and most accounting, CRM and email marketing platforms have added AI tools too. They're often switched off, half set up, or unknown to the people who'd benefit.
A general assistant such as ChatGPT, Claude, Gemini or Copilot covers a surprising amount of the list above. Choose one for the business, use its business tier so your data is handled properly, and learn it well. Five people using one tool properly will get more done than five people each dabbling with a different free one.
Run one small trial properly
Pick one task from your list and trial it for four to six weeks. Properly means you know roughly how long the task takes now, you've written down how the tool will be used, one person owns the trial, and you check the quality of the output as well as the time saved.
At the end you'll have a real answer, in the form 'this saves us around so many hours a week and the quality held up', or 'it didn't'. Both are useful. What you want to avoid is a vague sense that AI is probably helping while the subscription quietly renews every month.
Set a few rules before anyone pastes in customer data
For a small business, the biggest AI risk isn't a system going rogue. It's a well-meaning member of staff pasting a customer's details, a contract or the payroll spreadsheet into a free tool with unclear data terms.
You don't need a 20-page policy. A single page that says which tools are approved, what must never be pasted in, and who checks AI-written work before it reaches a customer will prevent most problems. If you handle personal data, and almost every business does, UK GDPR still applies once that data goes into an AI tool.
When it's worth building something custom
Off-the-shelf tools will take most small businesses a long way. Custom work starts to make sense when the same multi-step process happens many times a week and involves your own systems. An enquiry that has to be checked against stock, pricing and availability before a reply goes out, for example, or documents that need reading, classifying and filing into the right place in your CRM.
That's where AI connected to your own data and tools, with a person approving the important steps, can take hours out of the week. It's also where a poor design costs real money, so it's worth scoping properly before anything gets built.
A sensible first six weeks
None of this is glamorous, but it's how AI ends up saving time rather than adding another subscription to the pile.
- Week 1: list the repetitive, text-heavy, checkable tasks in the business
- Week 1: find out which AI features you already pay for
- Week 2: choose one approved assistant and write your one-page rules
- Weeks 2 to 6: trial one task, with one owner and a before-and-after measure
- Week 6: keep it, change it or drop it, then pick the next task
What's the best AI tool for a small business?
There isn't a single best tool. For writing, summarising and analysis, the main assistants (ChatGPT, Claude, Gemini and Microsoft Copilot) are all capable. Pick the one that fits the software you already use, pay for its business tier, and get good at it before adding anything else.
Will AI replace my staff?
For most small firms, not in the near term. The British Chambers of Commerce's 2026 research found that 95% of SMEs using AI reported no impact on the size of their workforce over the previous year. In practice, people tend to spend less time on admin and more on the work that needs them.
Do I need an AI policy?
Yes, but it can be short. One page covering approved tools, what must never be shared with them, and who checks AI-written work is enough for most small businesses to start with.
Not sure where AI would actually help your business?
Tell us how a normal week runs and we'll point to the two or three places AI is likely to pay off, and the places it won't. It's a free, no-pressure consultation, and we never chase you afterwards.
