Case study · Dental · 5 May 2026 · 8 min read

How a multi-practice dental group answers every patient call with one AI agent

An anonymised look at how a three-practice dental group in West England moved from three separate reception desks — each fielding NHS, private, and emergency calls in parallel — to a single AI receptionist that picks up every call on the first ring, across all three sites, 24/7.

All figures in this case study are anonymised at the client's request. We do not share practice names, addresses, or location-specific identifiers beyond "West England". Where percentages or counts are quoted, they reflect the client's reported figures from their first three months on the service. Method notes at the end.

The brief: three practices, three reception desks, missed calls

The group runs three high-street dental practices across West England, with a roughly even split of NHS and private patients. Across the three sites they had four front-of-house staff — two full-time and two part-time — covering practice hours of 8:30am to 5:30pm Monday to Friday. Like most UK dental practices, the busiest call window of the day was the first hour after opening, followed by a second spike around lunch when patients call between meetings.

The presenting problem wasn't that staff were unfriendly or untrained. It was simpler than that: there were more calls than ears. When the front desk at Site A had three patients checking in, the practice manager at Site B was processing a private treatment plan, and the part-timer at Site C had nipped to the kitchen, the phones rang out. Voicemails piled up. Some patients left a message; most didn't. Most just rang the next practice they could find on Google.

The brief from the practice principal was three lines on a Monday morning email: "We need to answer every call. We can't add more staff right now. We need a British voice that handles NHS as confidently as private." We scoped the work that Friday and went live the following Friday.

What we found in the first call audit

Before configuring anything, we asked the client for one week of inbound call logs from their VoIP provider (Yealink-based, going through a small UK SIP trunk). The numbers were uncomfortable. Across the three sites, the group received an average of 47 inbound calls per practice per day. Of those, the call data showed an answer rate of around 68% during opening hours and effectively zero outside them. About 22% of all unanswered calls came in between 5:30pm and 8pm — patients ringing on their way home from work to ask about an appointment they'd been putting off.

The lost-revenue maths writes itself. At an average new-patient lifetime value of roughly £600 for an NHS adult and £1,800+ for a typical private-leaning patient, even a 10% recovery on missed out-of-hours calls is meaningful. But the group wasn't running this as a maths problem. The principal was running it as a patient-experience problem: nobody who phones a dentist in pain should hear an engaged tone.

What Sofia handles each day

We named the agent Sofia at the client's request — a calm, professional British voice (Matilda, ElevenLabs) that callers consistently treat as a human receptionist unless they ask directly. Sofia answers the main number for all three practices on a Twilio bridge, with the same call flow logic but a per-site variable that determines which calendar she books into and which practice she names herself for. From the caller's side, ringing Site A's number sounds identical to ringing the human receptionist at Site A — same opening line, same warmth, same sign-off.

Sofia handles five categories of call across an average day, and we benchmarked her behaviour on each one before going live.

• NHS enquiries. She checks the NHS waiting list, takes the patient's details, explains banding (Band 1 / 2 / 3) when asked, and sends an SMS confirmation that they've been added to the list. For genuinely urgent NHS cases out of hours, she routes the caller to NHS 111 with a clear handoff line.
• Private bookings. She has the practice's private menu loaded — from a Band-2 equivalent private check-up through to implants and Six Month Smiles ranges. She quotes confidently, books a free consultation where appropriate, and sends a calendar invite the moment the call ends.
• Emergencies. She listens for severity markers (swelling, trauma, severe pain rating) and prioritises an in-hours emergency slot at the patient's nearest of the three sites. Out of hours, she routes to NHS 111 plus the on-call principal's mobile if the practice has one rota'd.
• Existing-patient admin. Reschedules, cancellations, recall confirmations, and "what time is my appointment" lookups. She speaks to the practice's calendar directly — no front-desk handoff needed.
• Review follow-ups. The next morning she sends an SMS with a Google review link to every patient she booked who actually attended. The result: fewer reminders for the front desk to chase, a steadier flow of 5-star reviews, and a measurable lift in the group's local pack ranking.

Results in the first three months

We don't quote exact percentages we can't fully attribute — that's a quick path to over-claiming. What follows are the figures the client reported back to us at the 30, 60, and 90-day reviews, with the caveats they came with.

• Answer rate climbed to ~99%. Sofia picks up every call within one ring. The remaining ~1% is calls that drop on the caller's side (signal loss, accidental dial) before she can respond.
• Out-of-hours bookings became a real channel. In month one, the group recorded roughly 60 new bookings per month coming in between 5:30pm and 9pm — calls that previously hit voicemail, of which the vast majority would never have rung back.
• No-shows down materially. The combination of confirmation at the booking call, SMS confirmation, and a 24-hour reminder dropped reported no-show rates by a band the principal described as "the difference between two empty chairs a week and roughly one a fortnight." Honest caveat: we can't separate the AI confirmation effect from a general SMS-reminder effect; some of this improvement would have come from any modern reminder system.
• Google reviews rose ~3x. The post-appointment SMS review nudge tripled the reported monthly review count across the group. The lift in local pack visibility on Google Maps is the second-order benefit they didn't pay for but quickly noticed.
• Front-desk staff stayed the same size, but did different work. The practice manager described the shift simply: "We're not running between phones and patients any more. The phone's handled. We do the patients in front of us."

What we learned (and what we'd change)

No deployment is clean. Three things genuinely surprised us, and one thing we'd do differently next time.

Surprise 1: NHS callers preferred Sofia to a human

We'd assumed older NHS patients might prefer a human voice. The opposite was true on average. Sofia doesn't rush, doesn't sigh, doesn't have a queue of three patients standing at the desk while she's on the phone, and explains NHS bands the same way every time. Several callers thanked her for being patient. One politely asked her name twice. She gave it twice.

Surprise 2: Pronunciation of clinician names took two iterations

The first week, Sofia mispronounced one of the partners' surnames (a non-English name with three vowels in a row). We added a pronunciation override in the prompt and re-tested across 30 sample calls before declaring it fixed. Lesson: every multi-clinician practice will have at least one name that needs a pronunciation override. Build it into the launch checklist.

Surprise 3: Emergency triage needed two passes

Our first triage flow over-routed to NHS 111. Patients with moderate but manageable pain were being sent to 111 when an in-hours emergency slot would have been the right answer. We retuned the severity thresholds in week two, with the principal calibrating each tier. Triage is the highest- stakes flow on the agent and deserves the most rounds of testing.

What we'd do differently

Start with a one-week shadow mode. We went straight from internal QA to live, which worked, but a week of Sofia answering only out-of-hours calls (with a human still picking up in-hours) would have built front-desk confidence faster. We now offer this as the default rollout for any group of two or more sites.

Want to see this in action?

The same playbook works for a single-site practice or a group of any size. The vertical-specific capabilities — NHS triage, private quoting, emergency routing, recall logic — are documented on the dental lander. We don't charge per minute, we don't charge per booking, and the setup is a one-time fee that includes call-flow design, British-voice calibration, and integration with your practice management system. See the three plans →

Or hear Sofia answer a real dental call right now: +44 1480 773536. It's a live demo agent — ask her about NHS appointments, implants, or an emergency, and she'll handle each like she handles them every day for the West England group.

Method notes. All figures are reported by the client at 30, 60, and 90-day reviews and have not been independently audited. The group's identity is withheld at their request; we use "a multi-practice dental group in West England" in all public references. Average call volume, answer rate, and no-show figures are rounded. Where we couldn't cleanly attribute an improvement to Sofia versus an adjacent change (e.g. SMS reminders), we've said so.