The short answer
Most small businesses with a website and steady enquiry volume will get measurable value from a custom AI chatbot — fewer questions repeated to the owner, leads captured outside hours, faster pre-qualification before a call. The exceptions are businesses with very low traffic (under 200 visits/month — chatbot doesn't get used enough), or businesses where every enquiry needs human nuance from the first message (high-end consultants, legal, specialist medical).
The honest "should you?" framework:
- Do you get the same 5-10 questions over and over? Yes → strong fit.
- Are you losing leads outside working hours? Yes → strong fit.
- Do customers want to chat rather than fill in a form? Yes → strong fit.
- Is your service highly nuanced from message #1? Yes → weak fit, keep humans.
- Is your site getting under 200 visits/month? Yes → fix the traffic first.
What changed
Chatbots got a bad name in 2018-2022 because they were rule-based — keyword matching to decision trees. They couldn't handle paraphrased questions, missed context, and felt robotic. That era is over.
Modern chatbots built on frontier LLMs (Claude, GPT, Gemini) are a different category. They can:
- Understand intent, not keywords
- Hold multi-turn context across a conversation
- Speak in a specific tone of voice consistently
- Refuse politely when something's out of scope
- Hand off to a human at exactly the right moment
The one on this site uses Claude Haiku 4.5 by Anthropic — fastest frontier-class model, costs pennies per conversation. The same approach I'd use for you.
Cost and ROI
A custom AI chatbot is £299 setup + £39/month. Over 12 months that's £767. Compare that to:
- The cost of a part-time receptionist (£800-£1,500/month)
- The cost of one lead lost because nobody answered the website at 9pm
- The cost of repeatedly explaining the same FAQ to every visitor who calls
For most service businesses with £30+ average lead value, ROI is comfortable from month one. (More on the AI chatbot service.)
Won't it make things up?
The #1 worry — fair enough. The mitigation has three layers:
- System prompt constraints. The bot is explicitly told not to invent prices, opening hours, or services. When unsure, it offers to take details and have a human follow up.
- Knowledge-base scoping. The bot only has the information you've explicitly given it. If a customer asks about something out of scope, it acknowledges it doesn't know and escalates.
- Model choice. Claude is one of the most honesty-aligned frontier models. It refuses politely and admits uncertainty far more readily than older systems.
Pre-launch, I test the bot against 50-100 likely real-world questions to make sure it behaves under pressure. Edge cases get added to the system prompt as they surface.
Keeping it on-brand
The bot's tone is the most overlooked part of the build. Three tone calibration steps:
- Tone audit. What words do you use? Formal or casual? Do you say "brilliant" or "great"? Do you swear (some clients do, in trades)?
- Reference responses. 10-15 sample exchanges written in the exact voice the bot should use.
- System-prompt anchoring. The system prompt explicitly references the tone and includes good/bad examples.
Done right, customers don't realise they're talking to an AI for the first few messages. Done wrong, it reads like ChatGPT trying to sell solar panels.
When NOT to add one
Honest negatives:
- Very low traffic. Under 200 unique visits/month, the bot won't get enough use to justify even £39/mo.
- Complex regulated services. Legal, medical, financial — first-line replies need a human or a heavily-supervised setup. I won't build these without legal sign-off.
- You enjoy the calls. Some solo founders love the first phone conversation — it's how they qualify and convert. If that's working, don't break it.
If you want one
Try the demo (bottom-right of this page) for ten minutes. Ask it about pricing, FAQs, anything you'd ask if you were a real visitor. That's the experience you'd offer your own visitors. If it's the right fit, the AI chatbot service page has the full detail.
The chatbots that work for small businesses aren't the ones that try to replace humans. They're the ones that answer the questions a human shouldn't have to answer for the 50th time.