I spent years talking clients out of chatbots, which is a strange thing for a developer to admit, since I was turning down paid work. But the old generation of chatbots deserved it. They were decision trees wearing a smiley face: press one for opening hours, press two for prices, press three to feel your soul leave your body. They frustrated visitors, answered almost nothing, and made otherwise serious businesses look cheap. A bad chatbot is worse than no chatbot, because it actively teaches your customers that contacting you is a waste of time.
That advice has expired. Modern AI chat is a different animal. It can read your actual website content, your service list, your prices, your FAQ, and answer real questions in plain language. It can hold a conversation that doesn't follow a script. And crucially, it knows when to stop: a well-built one hands off to a human the moment a question goes beyond what it knows, instead of improvising an answer. I've built a few of these now, for clinics and service businesses, and the pattern for when they pay off has become pretty clear.
They pay off when the same questions keep arriving after hours. Clinics get this constantly: someone sits down at 10pm, finally has time to sort out their health, and sends the same three questions every patient asks. Do you treat this, what does it roughly cost, when can I come in. If nobody answers until morning, some of those people have already booked elsewhere, because the clinic across town replied first. One client of mine was losing exactly these evening enquiries and didn't fully believe it until we counted. The bot now answers at 11pm, handles those three questions, collects a name and a number, and the lead is sitting in their CRM when the front desk arrives at breakfast. Nobody's job changed. The enquiries just stopped evaporating overnight.
The same logic holds for agencies, gyms, tutoring centers, and anyone selling something that needs a little explanation before someone commits. If your product needs zero explanation, or your entire sales process happens face to face, a bot has nothing to do. And if your traffic is tiny, be honest with yourself: a chatbot can't fix a website nobody visits. Fix the traffic first, and the money is better spent there.
On cost, the honest answer has two tiers. Off-the-shelf tools run roughly $50 to $500 a month depending on conversation volume and features, and for a simple use case they're fine. You'll spend an afternoon feeding them your content and tweaking the tone, and you'll live within their limits. A custom build costs more upfront, but then it's yours: trained on your content, wired directly into your booking system and your CRM, speaking your languages, and running for a fraction of the subscription price. Which route makes sense comes down to one number, which is what a missed lead costs you. A clinic where a new patient is worth hundreds of dollars does different math than a shop selling twenty-dollar products.
One warning, because it's the thing everyone worries about and they're right to: an AI bot that's allowed to guess will eventually promise a discount you don't offer or invent a service you don't provide. The fix is architectural, not hopeful. Constrain it to your real content, give it a clean handoff path to a human, and log every conversation so you can see what it actually says. That's the difference between a bot you trust and one you switch off after the first embarrassing screenshot.
My honest rule, the same one I give clients who ask: if you get more than a handful of enquiries a week and can't reliably answer them within the hour, the math almost always works. If not, spend the money on the website itself first.
Not sure which side of the math you're on? Ask me — I'll give you a straight answer, even if it's "don't build one."
