An AI receptionist for clinics and small businesses
What an AI phone agent does when nobody can pick up, how it works, where it needs a human, and what it costs.
Meer Habib
Senior Mobile Engineer · Chittagong
A physiotherapist is with a patient for forty-five minutes. The clinic phone rings twice during the session, and both calls go to voicemail. One of them books somewhere else.
That's the problem an AI receptionist solves, and it's a narrower, more useful thing than the phrase suggests. I built one as part of Ringo, an AI business phone for small teams, so here's what it does, how it works, and where it needs a human.
What it does
When nobody can pick up, the agent answers in a natural voice, in the business's name. On a typical call it:
- Answers the common questions: opening hours, location, prices, what to bring, whether you treat a given condition. From a short list the business writes and keeps up to date.
- Books, moves or cancels an appointment in the calendar or booking system the business already uses.
- Takes a message when the caller needs a person, and sends the team a summary instead of a voicemail to listen to.
- Hands over to a human for anything urgent or sensitive, by transferring the call or flagging it for a callback.
Every call leaves a recording (where the law allows it and the caller is told), a transcript and a short summary in one inbox, next to SMS and WhatsApp messages. The team reads three lines instead of listening to three minutes.
How it works
There's less magic in it than people expect. Four parts:
- A phone number on a telephony provider. Ringo uses Twilio Voice, with CallKit and PushKit in the app so calls ring like normal phone calls even when it's closed.
- A voice agent that listens, thinks and speaks in real time. Ringo's runs on ElevenLabs and LiveKit.
- Instructions and knowledge: what the business does, its prices and policies, what the agent may and may not do, and when to hand over.
- Tools: the actions it's allowed to take, such as checking availability, creating a booking or sending a summary. The agent only acts through these, which is what keeps it predictable.
Where it needs a human
An agent like this is good at the routine 80% of calls. The design work is in the other 20%:
- Opening hours
- Location
- Prices
- What to bring
- Book or move
- Take a message
- Medical advice
- Complaints
- Anything urgent
- It must know what it doesn't know. Medical advice, complaints and anything urgent go to a person, every time.
- It must say what it is. Callers are told they're speaking to an assistant, and can always ask for a human.
- Recording and privacy rules differ by country. Some places require consent before recording; health data has stricter rules again. These decisions are made with the business before launch, not after.
- Someone owns the answers. Prices and policies change. If nobody updates the knowledge, the agent confidently gives last month's answer.
What it costs
There are two parts. Building and setting it up for your business is a fixed-price project, usually a matter of weeks rather than months, because the hard parts (real-time voice, telephony, the inbox) are known problems. Running it has a usage cost per call minute from the phone and voice providers, which scales with your call volume. I estimate both for your actual volume before we start, so there's no surprise bill.
For most small clinics and service businesses, the comparison that matters isn't against a full-time receptionist. It's against the bookings lost to voicemail every week.
If your phone rings while everyone's busy, tell me about your business and I'll sketch how an assistant would handle your calls: what it answers, what it books, and when it hands over.
Building something like this?
Booking new projects for Q4. Replies within 24h.