Straight answer: a well-built AI chatbot can answer common customer questions, qualify leads, book appointments, and cut your response time from hours to seconds, and it will do that reliably for maybe 40 to 70 percent of the conversations that hit your website. The rest still need a human. Anyone telling you it replaces your team or writes your sales for you hasn’t run one long enough to watch it fail.
What a chatbot is built to do (five jobs, not fifty)
I get asked this constantly, usually by someone who’s just watched a demo where the bot sounds like a person and now assumes it can run their business. It can’t. But it can do a short list of jobs very well, and those jobs are worth real money.
- Answer the same 10 to 15 questions you get asked every single day. Opening hours, pricing bands, delivery times, whether you cover a certain postcode, what’s included in a package. If you’ve answered it more than 20 times by email this month, it belongs in a chatbot.
- Qualify a lead before a human wastes 20 minutes on it. Budget, timeline, company size, the basics. A decent flow can sort a tyre-kicker from a buyer in under two minutes.
- Book appointments without the back-and-forth email. Connect it to a calendar and it can offer real slots, confirm, and send a reminder.
- Capture the enquiry that would otherwise vanish at 11pm. Most small business websites lose enquiries overnight and at weekends because nobody’s watching the inbox. A chatbot is.
- Hand off cleanly to a human when it hits a wall. The good ones know when to stop pretending and say “let me get someone to call you.” The bad ones just keep guessing.
Notice what’s not on that list. Closing a complex sale. Handling a upset customer. Anything that needs judgement rather than pattern-matching. That’s the bit most sales pages for these tools quietly skip.
What happened when I put one on my own site
I put a chatbot on my own website in 2024, before most people had touched one, mostly because I write and speak about AI and I wasn’t going to keep doing that from the sidelines. First version was rubbish. I’d used a generic template, fed it my “About” page and a handful of FAQs, and left it. Within a week it was telling people I offered services I’d stopped doing four years earlier, because the training data included an old page my web person had forgotten to take down.
The fix wasn’t the tool, it was me. I sat down for an afternoon, wrote out the actual 12 questions people ask before they book a call with me, wrote the answers the way I’d say them out loud, and rebuilt the flow. After that it started doing something useful: pre-qualifying speaking enquiries by asking about audience size, date, and budget range before anyone got to me. That single change meant I stopped getting on calls with people who had a £500 budget for a keynote and started spending that time on the enquiries that were worth having.
The lesson that nobody wants to hear: a chatbot doesn’t fix a messy business, it just makes the mess move faster. If your service pages are unclear, your pricing is vague, or your team gives three different answers to the same question, the bot will do exactly the same thing, just at scale, and now it’s doing it 24 hours a day without anyone noticing until a customer complains.
The numbers that matter
Vendors love to quote resolution rates because they sound impressive in isolation. Intercom, for example, has publicly claimed its Fin AI agent resolves around half of support conversations without a human touching them. That’s a real, useful number, but it depends entirely on how narrow and well-documented your product or service is. A software company with a clean help centre will see numbers like that. A bespoke service business with fifty edge cases won’t, at least not at first.
From what I’ve seen with clients running these on small business websites, a more honest range looks like this:
- 20 to 30 percent resolution in the first month, while the flow is still rough
- 40 to 55 percent once you’ve fed it three months of real transcripts and fixed the gaps
- 60 to 70 percent is close to the ceiling for most service businesses, because the remaining questions need a person
On response time, the shift is more dramatic. A plumbing business I worked with in Kent used to average around four hours to reply to a website enquiry, because the office manager checked the inbox twice a day between jobs. A basic chatbot got that first response down to under a minute, which meant they stopped losing after-hours callouts to the competitor down the road who happened to answer the phone at 8pm. They estimated it recovered them roughly £1,800 a month in jobs that would previously have gone cold overnight. That’s not a huge business transformed by AI. That’s one leak plugged.
Where it falls down (and why that’s fine if you plan for it)
A chatbot cannot read tone. If a customer is furious, it will cheerfully keep offering the same three menu options while the person gets angrier. It cannot make a judgement call on a refund that sits outside your written policy. It cannot notice that a “quick question” is a customer about to leave you for a competitor. This is where the handoff to a human has to be built in, with a clear trigger, not an afterthought.
There’s also a trust problem nobody likes admitting. Data from our own AI Search Demand Report 2026, which looked at 3.4 million searches, showed a huge and growing appetite for “how do I use AI in my business” style queries, but also a persistent undercurrent of “will this annoy my customers.” It will, if it’s badly done. A chatbot that traps people in a loop, or that pretends to be human when asked directly, does damage to trust that takes far longer to repair than the time it saved.
And here’s the bit that gets left out of most guides on this topic: a chatbot is a marketing decision, not just a support tool. Every answer it gives is your brand voice, at scale, unsupervised, at 2am. If you wouldn’t let your newest, least-trained employee answer customers with no oversight, don’t let a badly configured bot do it either. This connects directly to how you use AI across marketing generally, which I’ve written about in more depth in how to use AI in marketing, because a chatbot is really just one channel of a much bigger shift in how customers expect to reach you.
What it costs, roughly
Pricing has settled down a bit since 2024’s land grab. As of 2026, for a small business, expect something like this:
- Basic scripted chatbot (no real AI, just decision trees): free to around £30/month, tools like Tidio or the free tier of most helpdesk platforms
- AI-powered chatbot trained on your content, connected to a calendar or CRM: roughly £50 to £300/month depending on volume, covers tools in the Chatbase, Intercom Fin, and ManyChat range
- Fully custom build with a developer, integrated into your booking system and CRM: £1,500 to £8,000 upfront, then £100 to £400/month in hosting and model costs
The build time matters more than the price. A basic version can be live in an afternoon. A version that’s trained on your real content, tested against real questions, and connected to your booking system typically takes two to four weeks of proper setup, including a week of just watching real conversations and fixing what’s wrong.
How to set one up without wasting a month on it
This is the part people skip and then wonder why the bot is useless.
- Pull your last 100 customer emails or enquiries. Not a guess, the actual messages. Sort them into buckets. You’ll almost always find that 60 to 70 percent of everything fits into fewer than 15 question types.
- Write the answers yourself, in your voice, before touching any software. If you skip this, the bot writes them for you, badly, using whatever it scraped from your website.
- Pick a tool that matches your actual traffic, not your ambition. If you get 40 website visitors a day, you don’t need an enterprise platform.
- Set a clear handoff rule. After two failed attempts to answer, or on any mention of “refund,” “complaint,” or “cancel,” it hands to a human, no exceptions.
- Read every transcript for the first two weeks. Not a summary, the actual transcripts. This is where you find out what people really ask, which is never quite what you assumed.
- Update the flow monthly, not once and forget it. A chatbot left untouched for six months quietly goes stale as your pricing, services, and policies move on without it.
If this whole process sounds like more than you want to take on alongside running the business, it’s exactly the kind of thing an AI consultant for small business is worth bringing in for, someone to do the messy transcript-reading and flow-building once rather than you learning it by trial and error over six months.
If you run a coaching, consulting, or agency business rather than a product business, the setup is a little different again, since your questions tend to be about process and fit rather than stock and delivery. I’ve gone into that in more detail in AI for service businesses, and it’s also worth reading what small business owners really do with an AI consultant if you’re weighing up doing this yourself versus paying someone to build it the first time.
The bit that decides whether this works for you
The businesses that get real value from a chatbot are the ones with a repetitive question load, an actual gap in after-hours coverage, and someone willing to check the transcripts. The ones that get nothing from it are the ones who installed it, walked away, and expected it to sell for them. A chatbot is a very good filter and a very poor salesperson. Use it for the first job and it’ll pay for itself within a couple of months. Use it for the second and you’ll be explaining to a confused customer why the bot promised them something you never offer.
None of this is complicated once you’ve seen it done, and if you want the wider picture of what’s changing week to week in this space, I cover the practical shifts as they happen in AI news this week for small business, and if any of the terminology in this post threw you, the plain-English breakdown in the AI marketing glossary is worth bookmarking.
Frequently asked questions
Can an AI chatbot replace my customer service team?
No, not fully. A well-built one can handle 40 to 70 percent of routine questions and free your team from the repetitive stuff, but complaints, refunds, and anything emotional or unusual still need a person. Treat it as a filter, not a replacement.
How long does it take to set up an AI chatbot ?
A basic version can go live in an afternoon. A version trained on your real content, tested against genuine customer questions, and connected to your booking system usually takes two to four weeks, including time spent reading actual conversations and fixing what’s wrong.
Will a chatbot increase my sales?
Indirectly, yes, by capturing enquiries you’d otherwise lose overnight and by qualifying leads so your team spends time on the right people. It rarely closes complex sales by itself. Expect it to protect revenue you’re currently leaking, not to generate brand new demand.
How much does an AI chatbot cost for a small business?
Basic scripted tools start free to around £30 a month. AI-powered chatbots trained on your own content and connected to a calendar or CRM typically run £50 to £300 a month. A fully custom build costs £1,500 to £8,000 upfront plus £100 to £400 a month in ongoing costs.
Useful references
Related reading: B2B Lead Generation Case Studies: Which AI Marketing Tools Paid Off and How AI Tools Can Help You Write a Research Paper Faster (Without Getting You Caught Out).
If you want the full breakdown, here is everything I know about AI marketing.
