Case study: an AI receptionist for Westbury Dental Practice
Pilot deployment in final testing before go-live. This page describes what has been built and tested. Measured results will be added once the practice has been live long enough to have them.
This is a pilot deployment built by Axivon's founder, Mags G, with Westbury Dental Practice. The AI receptionist answers the phone, checks who is calling, books routine appointments into the practice's own diary and hands everything else to a person. Below is what it does, how it was tested, and what comes next.
A busy front desk,
and a phone that keeps ringing.
A dental front desk greets patients, runs the diary and answers the phone at once. Calls that arrive while everyone is busy, or after closing, ring out or go to voicemail.
The practice chose a careful arrangement: reception's phones ring first, and the AI receptionist answers only when nobody picks up.
What the receptionist
actually does.
- 01
It opens with a fixed, pre-recorded notice telling the caller they are speaking to an automated receptionist, followed by a fixed greeting.
- 02
It asks whether the caller is a new or an existing patient.
- 03
If the calling number matches one patient record, the greeting can use their first name, but that is never treated as proof of identity. Before booking, an existing patient gives their date of birth and postcode, checked against the practice's live record. If the check fails, the call goes to reception.
- 04
For a verified existing patient, it looks at their usual dentist's diary first and asks before searching other dentists.
- 05
A new patient is offered an initial examination with the dentists the practice chose. It takes their name, date of birth, phone number, address and NHS or private status, reads them back and waits for the caller to confirm before anything is written.
- 06
Hygiene, cosmetic and implant appointments follow the practice's own rules on who sees them and when.
- 07
It answers questions on opening hours, prices and NHS cover using only facts the practice owner approved.
- 08
If a caller describes signs of a medical emergency, it gives fixed advice to call 999, go to A&E or call NHS 111, and books nothing. It never diagnoses, and never dials emergency services itself.
From call
to diary.
Built and tested against the Cloud4Dentists booking system. Other practice systems are assessed case by case.
Free times are worked out the way the front desk does it: each dentist's working pattern, minus appointments already in the diary, minus lunch. Slots are 30 minutes, need two hours' notice and run up to 60 days ahead.
It only says an appointment is booked once the booking system has confirmed it. A request that times out is flagged for a person to check, not sent again, so a slow response cannot become a double booking. Every request is logged in an encrypted access audit, and nothing is ever deleted through the system's interface.
What it hands
to a person.
Cancellations, changes to existing appointments, clinical questions and record requests go to reception.
When a call needs a person, it becomes a task in a staff inbox we built for the practice, with a call-reason category and a one-line summary. A callback is only saved once the caller has agreed to one, and a call that ends unresolved can also create a task.
During opening hours it can transfer the call to a separate practice line, telling the caller first. If nobody answers within 20 seconds, the caller is given the practice's number.
Paying before
the booking.
Status: in testing. The practice wants patients to pay before an appointment is confirmed.
During the call, the receptionist texts the caller a secure Stripe payment link and holds the slot for 30 minutes. The booking is written only after Stripe confirms payment, then a confirmation text follows. If someone pays and the slot can no longer be booked, reception gets an urgent task. It never takes card numbers over the phone.
Today this runs in Stripe's test mode for approved test numbers.
How it earned a place
near real callers.
This is the strongest evidence we have. Testing runs in three tiers, and none replaces another.
- 01
Every change runs the offline test suites, including end-to-end payment journeys that check the call never says "booked" and that an unpaid hold lapses. A failure blocks the change.
- 02
Before any release that changes what callers hear, every model-driven scenario runs: 60 regular call scenarios and 36 payment scenarios. An AI caller talks to the receptionist's real instructions and a separate model judges each conversation. Every failure is classified as a receptionist bug, a test fault or a judging error, and the test is fixed rather than the expectation lowered.
- 03
Before real payments or any change to the phone path, Mags makes supervised calls from approved test numbers. They write real test appointments into the live diary, which are then removed by hand. Only approved test numbers can reach the live booking path.
What callers are told,
and what is kept.
The practice is the controller of its patients' data and Axivon is its processor. Callers are told at the start that they are speaking to an automated receptionist.
For ordinary calls no transcript is kept, and reporting holds counts, not patient details. Recognising a caller's number uses one-way coded keys rather than a copy of the patient records.
Only calls from approved test numbers are recorded, after the notice has played. Recordings sit in private UK storage and are deleted after 30 days. Recording ordinary callers would need separate approval.
A data processing agreement and a data protection impact assessment have been drafted and reviewed by a solicitor.
Before patients
reach it.
- 01
The practice switches on call forwarding, so the receptionist picks up when the front desk cannot.
- 02
Public booking is switched on in one step, once everything has been tested together.
- 03
Live payments follow, using the same secure payment link that has been tested.
Once the practice has been live long enough to have them, we will add measured results here.
Book your
demo.
Tell us how work comes in and where it hurts. We'll tell you honestly whether we can help — no pitch, no cost.