What an AI chatbot can really do for a small business
A plain look at what modern AI assistants do well, where they must stop, and a quick way to tell whether your business needs one yet.
You've seen the chat bubble in the corner of other websites and wondered whether your business needs one. Maybe a software rep has told you AI will answer every enquiry while you sleep. The truth is more useful and a little less exciting: a good AI assistant handles the repetitive first part of a conversation well, and it needs clear limits to do that safely.
What is an AI chatbot, compared to the old kind?
The chatbots most people remember were decision trees. You clicked "Pricing" or "Opening hours", it showed a canned reply, and anything off-script hit a dead end.
A modern AI assistant is built on a large language model, like the ones behind Claude or ChatGPT. It reads what the visitor actually typed, in their own words, and answers from information you've given it about your business. It copes with spelling mistakes, follow-up questions and someone asking three things in one message.
That flexibility is the point, and it's also the risk. A model that can phrase anything can also phrase something wrong with complete confidence. So the real work in setting one up isn't the chat window. It's deciding what it knows, what it's allowed to say and when it must stop and hand over.
What a good AI assistant can do
For most small service businesses, the useful jobs fall into four groups.
- Answer. Service areas, opening hours, what you do and don't do, how the process works, what to expect on the day. The questions you answer ten times a week.
- Qualify. Ask a few questions before a person gets involved: what job, which suburb, how soon, roughly what budget. A plumber in Moreleta Park doesn't need to phone back someone in Polokwane.
- Book or capture. Offer times from your real calendar, or take a name, number and short description and push it into your CRM, so nothing lives only in a chat log.
- Hand over. Pass the conversation to you, with a short summary, when it's urgent, complicated or the person simply wants a human.
The hand-over is the part most cheap setups skip, and it's the part that decides whether customers trust the thing.
What an AI chatbot shouldn't do
This is where I'm strict when I build these.
- It shouldn't invent prices. If your prices are fixed and published, it can quote them. If every job is quoted after a site visit, it says so and offers to book one. It never guesses a number to be helpful.
- It shouldn't promise outcomes. No "we'll be there within the hour" unless that's a rule you've actually given it.
- It shouldn't wander outside its lane. A salon assistant in Menlyn can explain treatments and prices. It shouldn't be diagnosing a skin condition.
- It shouldn't pretend to be a person. Say it's an assistant. People are fine with that when it's quick and useful.
- It shouldn't collect more personal information than it needs. A name, number and job description is usually enough. Under POPIA you should be able to say why you're collecting each piece.
The safety net behind all of this is an "I don't know" rule. When the answer isn't in the information you gave it, the assistant says it isn't sure and offers to get you involved. I go through how to write those rules in how to train an AI assistant on your business.
How to tell if your business needs one
An AI assistant earns its keep when there's a gap between when people ask and when you can answer. A quick self-check:
| If this sounds like you | What it suggests |
|---|---|
| Lots of enquiries after hours or on weekends | An assistant can reply and capture details while you're off |
| The same five questions every day | Good fit: it can answer those from your own information |
| Leads go cold because you reply hours later | It can make the first reply instant and alert you to the good ones |
| Every job is different and needs a site visit | It can still qualify and book, but it shouldn't quote |
| Only a handful of enquiries a week | You probably don't need it yet. Answer them yourself |
| Most enquiries come by phone, not chat | Look at phone automations first |
If you're in the last two rows, save your money. A clear website with a WhatsApp button and quick replies from you will do more than a bot nobody uses. For phone-heavy trades, missed-call text back is often a better first automation.
Website chat, WhatsApp or both?
Most people start with a chat widget on the website because it's the simplest to set up and test. In South Africa, though, plenty of customers would rather message you on WhatsApp. An assistant on WhatsApp is possible, but it needs the WhatsApp Business Platform rather than the normal app, and Meta charges for some messages.
Start where your enquiries already come in. If you're not sure, count last month's: how many came from your website form, how many from WhatsApp and how many by phone. That tells you where an assistant would actually be used.
What an AI chatbot costs, roughly
There are three layers. The setup covers writing its brief, connecting it to your calendar and CRM, and testing. A monthly fee covers keeping it maintained and improving it. Then there's usage: what the AI provider charges for each conversation. For a small business, usage is usually modest, but it grows with volume.
My website assistant is R6,500 setup plus R1,950/month, with AI usage billed at cost. I go through the full picture, including cheaper do-it-yourself options, in what an AI chatbot costs in South Africa.
Before you switch one on
Whoever builds it, insist on these five things:
- A written brief for the assistant: what it knows, what it can say and what it must never say.
- Testing with real enquiries you've had in the past, not made-up ones.
- A clear hand-over path to a person, tested end to end.
- Every conversation logged somewhere you can read it.
- A weekly read of those logs for the first month, fixing anything it got wrong.
That last one matters most. An assistant gets good because someone reads what it said and tightens the brief. It doesn't improve on its own, and nobody should tell you it does.
Where I can help
I build AI assistants for small businesses that answer from your own information, qualify, book and hand over to you, connected to your CRM so every lead lands in one place. If you're not sure you need one, tell me how your enquiries come in and I'll tell you straight whether it's worth it.
Questions people also ask
Are there any hidden fees?
No. Every service page lists exactly what you pay once-off and what runs monthly (like the care plan at R350/m for hosting, security and small updates). What you see is what you pay.
Can I mix services?
That's the whole point. Most clients start with a website, then add automation, ads or video as they grow — everything is built to connect.
When can I cancel monthly services?
Anytime. Send a message saying you'd like to cancel and your subscription is closed within 72 hours — no lock-in contracts.
Do you build online stores?
Yes — e-commerce builds from 30 to 200+ products with payment gateways, reviews and abandoned-cart automation. Ask about it on the Website Design page or in the form below.
Do you work with overseas clients?
Yes — I'm based in Pretoria, South Africa, and work with select UK and US clients remotely. Time zones have never been a problem.
Will customers mind talking to a bot?
Most people mind slow replies more than they mind a bot. What annoys them is a bot that pretends to be human, goes round in circles or won't let them reach a person. Say upfront that it's an assistant, keep answers short, and make reaching you one tap away.
Can an AI chatbot reply in Afrikaans or other languages?
The big models handle many languages, including Afrikaans, but quality varies from language to language. If a lot of your customers write in something other than English, test the assistant with real messages in that language before you rely on it, and keep the hand-over to a person easy.
Do I need a CRM for an AI chatbot to be useful?
Not strictly, but it helps a lot. Without one, leads sit in chat logs and emails and get forgotten. With a CRM, every conversation becomes a contact with notes, a stage and a follow-up, so you can see what came in and what happened next.