AI Voice Agent for Private Clinic Bookings
The clinic was losing up to 30% of patients due to busy phone lines at peak hours. We deployed a voice AI agent that answers calls 24/7, books appointments, and sends SMS confirmations.
- −30%missed calls
- 94%successful bookings
Industry context
In private healthcare the phone is still the main booking channel, especially for older patients and when the question is urgent. Yet calls are distributed very unevenly: Monday morning and the first hour after lunch create a peak that two receptionists physically cannot absorb. A patient who does not get through almost never calls back. They dial the next clinic on the list. So a missed call here equals a lost patient, and the cost is measured not in one appointment but in the whole lifetime of a client who went to a competitor.
The challenge
At peak hours the front desk could not answer every call, and patients simply went to competitors. Bookings were manual, so double-bookings and missed confirmations happened.
Our approach
- 01
Call script
We scripted the dialogue: greeting, service selection, free slots, confirmation.
- 02
Calendar integration
We connected the agent to the doctors’ schedule for real-time booking.
- 03
SMS confirmations
We added automatic reminders and confirmations to cut no-shows.
How the solution works
- 01PatientCalls the clinic
- 02Bland.ai + ElevenLabsThe agent handles the call by voice
- 03n8n + Google CalendarBooks a free slot
- 04SMSSends a confirmation
- 05Front deskTakes the complex requests
ElevenLabs handles the voice, Bland.ai runs the telephony and conversation logic, and n8n ties booking into Google Calendar and sends SMS confirmations. Complex requests are handed off to the front desk gracefully.
Why this stack
Bland.ai runs the telephony and conversation logic, because a voice channel imposes a requirement chat does not: latency. A pause over a second reads as a dropped call and the person hangs up, so response speed took priority over maximum model sophistication. ElevenLabs handles synthesis: in a medical context a robotic voice undermines trust faster than any scripting error. n8n ties booking to the doctors’ real schedule. This is a critical part, because an agent offering an already-taken slot is worse than an answering machine. Google Calendar stayed the source of truth for scheduling so receptionists could keep working in their familiar tool without a parallel system.
Where projects like this usually break
An agent that tries to treat
The most dangerous boundary in a medical voice agent is advice. A patient will inevitably ask “what could this be.” The agent must recognize such questions and hand off to a human rather than answer. This is designed into the script from the start, not added after an incident.
Recognizing names and addresses
Ukrainian surnames, street names and phone numbers are the weak spot of any voice agent. Without mandatory spoken confirmation and an SMS duplicate, a share of bookings will be for people who do not exist. Confirmation has to be part of the script, not an option.
A schedule that drifts
If a receptionist books a patient into a paper log or a separate system, the agent does not know and sells the same slot twice. Before launching voice booking the schedule must live in one place, otherwise automation creates more conflicts than it solves.
Results
| Before | After |
|---|---|
| Calls went unanswered at peak hours | 30% fewer missed calls |
| Manual bookings with double-ups | Bookings go straight into Google Calendar, 94% succeed |
| Confirmations got missed | An SMS confirmation goes out automatically |
Inbound-call pilot within 2-3 weeks.
Who this fits
A voice agent pays off where there are measurable missed calls and a standard booking scenario: clinics, dental and veterinary practices, salons, service centres. The test is simple: look at your phone system statistics for how many calls go unanswered at peak hours. If it is dozens a week, that is real money. If instead every conversation is a complex consultation involving doctor selection and symptoms, only the first step is worth automating: take the call, capture the contact, hand off to a human. One prerequisite is non-negotiable: a digital schedule. Without it the project starts not with an agent but with fixing how bookings are recorded.
Frequently asked
Will the patient realize they are talking to a bot?
Some will, and that is fine: the agent introduces itself as the clinic’s automated assistant. Hiding it is wrong both ethically and practically: when a person knows they are talking to a system, they phrase things more clearly and the booking goes faster.
Does the agent give medical advice?
No, and that is hard-coded into the script. The agent books appointments and states opening hours, prices and how to prepare for a procedure. Any question about symptoms, diagnosis or treatment is a trigger to transfer to a receptionist or doctor.
What if the agent does not understand the patient?
After two failed clarification attempts the call transfers to a receptionist, and if they are busy the agent captures the number for a callback. The core rule: the patient must never be left with nothing, even when the script fails.
Does the agent work in Ukrainian?
Yes, Ukrainian is the primary scripted language. The voice is selected separately: in a medical context intonation and pace affect trust as much as content, so the voice is tested on real recordings before launch.