Two reminder windows. One bilingual AI receptionist. A practice manager who refused to add another login. The actual playbook from a real rollout.
Summit Dental is a four-chair general practice in a metro suburb. Six staff, including the dentist-owner. They book about 220 appointments a week and, before they came to us, were losing about 19% of them to no-shows. The industry average for general dental is 12-15%. Summit was 4 to 7 points worse than that, and it had been slowly getting worse for two years.
The practice manager, Lisa, had been hired 14 months earlier to fix exactly this problem. She had tried two reminder systems in that time. Both had been technically fine and operationally hostile — they added a separate login, separate notifications, separate manual reconciliation against the practice management software, and a meaningful amount of staff time chasing failed text messages.
When we first met, Lisa told us — and we wrote it down — "if this requires me to add another login to my morning, I am not going to do it. I have nine logins. I will pay you not to give me a tenth one."
The instinct is to assume no-shows are about forgetfulness. They are not, mostly. Across the 380-odd no-show calls Summit logged in the 90 days before our rollout, the post-hoc reasons broke down roughly:
The two categories that a good reminder system can convert — genuine scheduling conflicts and confusion about the time — together account for over half of no-shows. The "decided not to come" group will not be converted by a reminder; they will be politely confirmed and choose not to come anyway. The forgetful 23% are the easy wins.
A reminder system that only handles "forgetful" will capture maybe a quarter of the value. A reminder system that handles "forgetful + scheduling conflict + confusion" — and converts the conflict-driven cancellations into rescheduled appointments rather than empty chairs — captures most of it.
We landed on a two-window reminder cadence. The first reminder runs 72 hours before the appointment and is informational — a confirmation call from the AI receptionist that confirms the time, the chair, and the procedure. The second reminder runs 24 hours before and is conversational — the AI is empowered to reschedule, transfer to a human, or cancel-and-rebook on the same call.
The two-window cadence matters. Most reschedules in the original data were customers calling 48-72 hours out, when they realised the appointment did not work and they still had time to fix it. A 24-hour single reminder catches them too late. Two windows give the conflict-driven cancellations a clean off-ramp into a rescheduled slot.
The AI receptionist runs on the same number patients had been calling for years. That was a non-negotiable. Asking patients to remember a new number, or to interact with a separate texting channel, would have lost most of the benefit. The phone number stayed the same; behind the number, the answering behaviour changed.
Lisa's nine-logins constraint was the design challenge that turned out to be the most useful. The AI receptionist had to read appointments out of Summit's practice management software (DentMaster), write reschedules back into it, and let Lisa see what was happening — all without adding a separate workflow for her.
We solved this in two ways. First, the AI writes directly into DentMaster's appointment table via their published API. When a patient reschedules a Tuesday 2pm to a Thursday 10am, DentMaster reflects the change within 30 seconds, exactly as if Lisa had typed it. There is no separate "Ajoxi calendar" she has to reconcile.
Second, the daily summary lands in Lisa's email inbox at 6:45am each morning — the same place her overnight messages have always landed — with five lines: "Yesterday's reminder calls. 32 confirmed. 4 rescheduled. 2 cancelled. 1 transferred to you. Details below." She does not have to log into anything to see it. If she wants the details, the email has a link.
The system is not actually login-less. There is an admin console where Lisa can change reminder timing, listen to recordings, or override AI behaviour for a specific patient. She has used it twice in six weeks. Her staff has used it zero times.
Roughly 18% of Summit's patient base speaks Spanish at home. Before the rollout, reminders had been English-only — because the previous systems either did not support Spanish or required Lisa to manually flag which patients needed Spanish reminders, which she did not have time to do.
The new system detects the patient's preferred language from DentMaster's patient-record field, where it is captured at first visit. If the field is set to Spanish, the AI calls in Spanish, end to end. If the field is blank, the AI opens in English but switches to Spanish within two seconds if the patient responds in Spanish. The detection is fast enough that the patient cannot tell from the audio that there was a language decision at all.
The bilingual coverage matters more than it sounds. Within the 18% Spanish-speaking subset, no-show rates had been roughly 4 percentage points higher than the practice average before rollout — almost certainly because English-only reminders were converting at a lower rate. After rollout, the Spanish-speaking subset has the same no-show rate as everyone else. The gap closed.
In the six weeks after the new system went live:
The chair utilisation number is the one that maps directly to revenue. A 7-point improvement on 220 weekly appointments, at Summit's average chair revenue, was meaningful enough that the practice's monthly billings rose by an amount that paid for the AI receptionist roughly twelve times over.
Two things that we tried and removed.
We tried a third reminder window two hours before the appointment for the highest-risk patients. It backfired. Patients reported the third reminder as annoying — three calls in three days felt aggressive — and the lift on the no-show rate was negligible because the people who were going to no-show at two hours had already decided. We removed it after ten days.
We tried letting the AI offer a small discount ($15 off the visit) to patients who looked at risk of cancellation. The dentist-owner caught this in week two and asked us to turn it off — not because it didn't work, but because she did not want a discount-conditioned customer base. We turned it off the next day. The result is part of the playbook: the AI is a tool the practice runs. It is not an autonomous discount engine.
AI receptionists, wholesale routes, virtual numbers — built on one platform with transparent pricing and a 24/7 NOC.