A voice agent for dental clinics that spots acute pain and books patients straight into the schedule
Patients call or write to the clinic with the same complaint: "my tooth hurts". The agent asks them eight questions, works out how soon they need a dentist and books them into the software the clinic uses for its schedule: IDENT, Dentrix, Open Dental or any other system with an API. The clinical director hears about the most urgent cases straight away on Telegram.
The clinical director answers WhatsApp at 23:40
Let's take a private dental clinic with two treatment rooms as an example. In 60% of enquiries, the patient complains of severe pain and asks for help right away. The rest are routine: cleaning, check-ups, cosmetic work.
Sorting mistakes are costly either way. If acute pain is missed, the patient is left without help and the clinic loses them. If a routine case is taken for an urgent one, it fills a slot that a patient in acute pain needs.
"My tooth hurts" says very little. The same words come from someone whose tooth ached after a cold drink and from someone with a fever and a swollen cheek. Telling the two apart is hard for the daytime receptionist and for whoever answers in the evenings and at weekends.
So the clinical director reads every message and decides who comes in. By my estimate, that takes 18 hours a week, half of them in the evenings and at weekends.
The agent asks the patient about the pain and books the appointment itself
A patient calls the clinic or sends it a WhatsApp message. The clinic keeps its phone number and has its phone provider forward calls to the agent. The agent asks eight questions about the pain and uses the answers to decide how soon the patient needs a dentist: straight away, within 24 hours or at a routine appointment. Then it books the patient into the clinic's scheduling system. For the most urgent cases, the clinical director gets a Telegram message at once.
Vapi or Bland AI handles the call; ElevenLabs provides the voice. On WhatsApp, the agent has the same conversation in text
An AI model, GPT-4o or Claude, asks them from a fixed list
From the answers, the model suggests the urgency: straight away, within 24 hours or routine. If there are warning signs, fixed rules decide
A workflow in n8n picks a time that matches the urgency and checks that it is free
The clinical director hears about the most urgent cases on Telegram at once. Other patients get an SMS confirming their booking
What the agent asks
The agent follows a fixed list of eight questions. It asks what the pain is like (sharp, dull or throbbing), how long it has lasted and where it spreads: to the jaw, the temple or the ear. It checks for fever and swelling and asks how the tooth reacts to cold, heat and pressure. The last questions are whether the tooth has been treated before and what medication the patient has taken.
The agent records the answers in the same format every time. They are used to set the urgency and are added to the patient's record in the clinic's system. The agent decides only one thing: how soon the patient needs to see a dentist. The diagnosis is made by the dentist at the appointment.
Warning signs are handled by fixed rules
The most dangerous signs are checked by a set of fixed rules, and the AI model plays no part in that check. If the temperature is above 38 °C and there is swelling, the patient is marked as most urgent straight away. The same happens if they can't open their mouth or have injured a tooth in the last 6 hours. The model can't change this decision, so it can't put such a patient in the general queue by mistake.
When there is no free slot
Bookings are made by a workflow in n8n, a tool that connects services to each other. It works with IDENT or any other practice management system that has an API, such as Dentrix or Open Dental. The workflow picks the room, the dentist and the nearest time that suits the urgency. If there is no free slot for an urgent patient, the clinical director gets all of the patient's answers on Telegram. The director then decides whether to move a routine patient, open an extra slot or refer the patient to an emergency dental clinic.
Voice agent platform: answers the call and recognizes the patient's speech
Gives the agent its voice: reads the replies aloud
AI model: asks the questions and suggests the urgency. It never receives the patient's name or phone number
IDENT or another system with an API. The agent uses it to book the patient into the right room at a free time
Links the steps: urgency, booking, message to the director. Runs on a server in the region required by local law
Sends urgent cases to the clinical director at once, with a short summary of the answers
Database on a server in the region required by local law, for example in the EU. It stores patient data and records which conversation belongs to which patient
The agent asks for consent to recording and data processing, as local law requires, and saves it in a log
Warning signs are handled by rules, and the AI model can't override them
Where patient data is stored and what the AI model sees
Patients' names, phone numbers and medical details are stored in the clinic's own PostgreSQL database. I set up storage to comply with local data protection law. For example, a clinic in the EU gets a server in an EU region. The clinic can pick the hosting provider itself: I work with any VPS.
The AI model receives only the age group, sex and symptoms, for example "45–55, male, throbbing pain for the second day". The model does not receive names, phone numbers or passport details. Only the clinic's database knows which conversation belongs to which patient.
At the start of the call, the agent asks for consent to recording and data processing, as local law requires. Every consent is saved in a log.
Call recordings are saved to the clinic's server. Before launch, I check where and for how long the voice platform keeps recordings on its side, and set up storage to comply with local law. I go over the wording of the consent with the clinic's lawyer in advance.
What to expect: an estimate for a typical clinic
All figures below are an estimate for the clinic in the example, based on typical numbers for small and mid-sized businesses in Russia. Each card shows how things usually are today, then the target after launch. For your clinic, I'll recalculate the figures during the audit, based on the enquiries you actually receive.
the share of emergency slots that go to patients who really need them
from the first call or message to a confirmed booking
only urgent cases and unclear answers are left for the director
estimate: about half of the 16 hours the director gets back fall within working hours. If the director spends them seeing patients, that's one more day of appointments a week
The main goal here is medical. A patient in acute pain is booked into the nearest free slot within minutes. A routine patient picks a time that suits them, and nobody has to wait for the receptionist to check and call back.
There's one more benefit. The report shows on which days and at what times urgent and routine patients tend to get in touch. The clinical director uses it to decide when to open extra emergency slots and when to schedule staff meetings and cosmetic work.
How long the launch takes
from the start of work to launch. The exact time depends on how easy it is to connect to the clinic's booking system
off the price list, plus a 90-day warranty
for 5,000 ₽: I'll look at how the clinic handles enquiries today and recalculate the figures for your clinic
Prices for voice agents are on the Pricing page.
Where else an agent like this helps
The idea is the same everywhere: ask detailed questions, work out how urgent it is, make a booking and pass anything urgent to a person straight away. It suits any business where urgent and routine requests arrive mixed together and a mistake is costly:
- → Multi-specialty clinics — the agent uses the symptoms to suggest which specialist to book
- → Veterinary clinics — an animal is unwell right now, or it's time for a routine vaccination
- → Service centers — equipment has broken down and needs an urgent call-out, or it's due for routine maintenance
- → Legal advice — a deadline is close, or the question is a routine one
- → Technical support for businesses — "the system is down" and "we need a new feature" call for different response times
- A ready list of questions for the agent to follow, with the answers saved in the same format every time
- Fixed rules that decide critical cases with no input from the AI model
- Personal data stored where local law requires; the AI model gets only what it needs to make a decision
- Urgent cases go straight to a person in Telegram or Slack, with a short summary of the conversation and buttons for the next step
If urgent and routine requests arrive mixed together and someone sorts them by hand, an agent can take over this work
When medical or other personal data is involved, I plan in advance where it is stored and what the AI model receives. Launch takes 5–7 weeks, depending on how hard it is to connect to your CRM or medical software.
Blog article on this topic: Personal data and AI: what reaches the model and how to check it
Related cases and ready-made solutions
Reviews on Yandex Maps and 2GIS: the shop asks every customer for a review, and AI helps draft the replies
Eighteen hours after a work order is closed in Alpha-Auto (1C), every customer gets the same WhatsApp…
After a missed call, the customer gets an SMS with the real ship date
The phone system reports a missed call, and the service finds the customer and their orders in two Shopify…
A store's own AI sales assistant won a blind test 34 : 12
Built for the brand's Russian site on Russian language models, so visitor data stays in Russia. It answers…
Start with the 5,000 ₽ audit: a written report and a payback estimate
I'll tell you what to automate first, when it will pay for itself and whether your task needs AI at all. Sometimes it doesn't.
Or just ask a question. During business hours I reply within 2 hours