Picture a small vet clinic at 7pm on a Saturday. Someone's dog has just eaten a bar of chocolate. They call, it rings out, and they call the next clinic on Google instead. The first clinic never finds out it lost that client.
That's the problem an AI receptionist solves. It doesn't replace your front desk. It catches the calls your front desk can't get to.
What happens on a normal call
Here's what a good setup does when someone calls a clinic at 9pm:
- 1Picks up within a ring
- 2Asks why they're calling
- 3Answers from your approved notes
- 4Checks your live calendar
- 5Books and texts a confirmation
The next day, you see a short summary of the call in your sheet or CRM: who called, what they wanted and what was booked. If they need a reminder, it goes out 24 hours before the appointment.
None of that is clever on its own. What makes it valuable is that it happens every single time, including on a Sunday.
The part most people skip: triage
The most important design decision isn't the voice or the script. It's deciding which calls the AI should not handle.
- A vet clinic: "My dog ate chocolate" goes straight to the on-call vet. "I'd like to book a vaccination" gets booked.
- A law firm: the AI collects the basics (what happened, when, where) and books a consultation. Anything urgent gets flagged to a person right away.
- A dental clinic: severe pain or swelling gets an urgent callback. A cleaning appointment gets booked.
Where it falls short
I'd rather you hear this from me than find out in week two:
- Bad lines and heavy background noise still trip it up sometimes. It should ask the caller to repeat, not guess.
- Upset callers usually want a human. A good setup notices that and offers a callback.
- Complicated pricing is risky. If your quote depends on an inspection, the AI should book the inspection, not quote a number.
- Anything not in its notes. It should say "I'll get someone to call you back about that", not make something up.
Is your business ready for one?
- You miss calls while you're with customers.
- People call outside your opening hours.
- Your front desk answers the same five questions all day.
- You don't actually know how many calls you miss each week.
If two or more of those sound familiar, it's worth a look. Each recovered call is a customer you'd otherwise have lost without knowing.
How I'd roll it out
Week one: after-hours and weekends only, so nothing changes during the day. Week two: read the transcripts together and fix the answers that were vague. After that: let it catch calls your team can't pick up during busy hours. You keep your number the whole time; calls are forwarded when nobody answers.
Questions people ask me
Will callers know they're talking to an AI?
They should, and a good setup says so in the first sentence. In my experience people care much more that someone picked up than about who it was.
Can I keep my existing phone number?
Yes. Your provider forwards calls to it when nobody answers or the line is busy. Your number doesn't change.
How much does it cost?
It depends mostly on call volume and how many systems it needs to connect to. On a call I'll look at your numbers and give you a fixed quote.
More on how we set this up: AI receptionist.
Founder of Overtake. I was a founding engineer at four startups, two of them in Y Combinator. I write about the AI systems I build for small businesses: what works, what doesn't, and how to start.
About meFollow on X