AI agents for UK small businesses: what actually works today
A plain-English look at the kinds of AI agent on offer in the UK, what each is genuinely good at, and the questions to ask before you pay for one.
First, what “agent” really means
Almost every software company now calls its product an AI agent. In practice the word covers very different things, from a chat box that answers questions about your website to a system that reads your inbox and books appointments on its own.
A useful test: an agent is something that takes an action for you, not just something that writes text. If it can’t send, book, update or file anything, it’s an assistant. Assistants are still useful, but they’re cheaper and lower-risk, and you should judge them differently.
The five kinds you’ll be offered
| Kind | Typical examples | Good at | Watch out for |
|---|---|---|---|
| Office assistants | Microsoft 365 Copilot, Gemini in Google Workspace, ChatGPT Business | Drafting, summarising email and meetings, finding things in your files | Only as good as how tidy your files are. Check who in the team can see what. |
| Customer service agents | Intercom Fin, Zendesk AI agents, Tidio Lyro | Answering repeat questions on your site or inbox, around the clock | Wrong answers delivered confidently. Needs a clear handover to a person. |
| Phone and booking agents | AI receptionists offered by telephony and booking providers | Answering missed calls, taking messages, booking simple appointments | Callers who dislike talking to a machine. Call recording rules. |
| Workflow automation with AI steps | Zapier, Make, n8n | Moving information between tools you already use, sorting and tagging | Quietly breaks when one of your tools changes. Someone has to own it. |
| Built-in features of tools you already pay for | AI features in Xero, QuickBooks, HubSpot, Shopify and similar | Small wins with no new contract or data sharing | Easy to overlook. Often the best place to start. |
Products and prices change monthly, so treat these as examples of each kind rather than recommendations. Check current UK pricing and terms before committing.
Where small businesses are getting real value
- Answering the same twenty customer questions, with a person picking up anything unusual.
- Turning calls, meetings and voice notes into written notes and follow-up tasks.
- Drafting quotes, proposals and replies that a person checks before they go out.
- Sorting incoming email and enquiries so the urgent ones surface first.
- Catching missed calls out of hours and taking a proper message.
Where it tends to disappoint
- Anything where a wrong answer costs money or trust, such as pricing, legal or medical advice, with nobody checking.
- Processes that aren’t written down. If your team does it differently every time, an agent will too.
- Replacing a relationship. Customers who chose you for the personal touch notice quickly.
UK-specific questions to ask every supplier
- Where is our data stored and processed, and will you sign a data processing agreement?
- Is our data used to train your models? Can we switch that off?
- How do we tell customers they’re talking to AI, and how do they reach a person?
- If it records calls or reads messages, what do we need to tell people under UK GDPR?
- What does it cost when usage grows, not just in the first month?
- How do we export our data and leave?
The Information Commissioner’s Office publishes guidance on AI and data protection that is worth a skim before you choose.
A sensible way to start
Pick one job that happens often, is low-risk if it goes wrong, and is easy to check. Try the cheapest option that could do it, ideally a feature in a tool you already use, for a month. Measure the time it saves against the time spent fixing its mistakes. Only then look at anything bigger.