An AI dental receptionist that books straight into Dentrix, Open Dental, Curve and Eaglesoft
Answering the phone is the easy part. Writing a confirmed appointment into a practice management system that was never designed to be written to is where most dental voice AI quietly gives up and takes a message instead.
30–40% of calls to a dental practice go unanswered. Each one dials the next practice.
The structural problem in a dental practice is not that the front desk is bad at answering phones. It is that the front desk is chairside with a patient, or on lunch, or already on a call when the second line rings. Industry baselines and our own client measurements put the missed-call rate between 30% and 40% at a typical practice. The dental network we now run was answering about 62% of its inbound calls before deployment.
Voicemail does not recover those calls, because most patients do not leave one — they call the next practice in the search results. Answering services do not recover them either: the quality varies shift to shift, the operator is reading a script with no view of your schedule, and the HIPAA liability still lands on you. Traditional IVR makes it measurably worse by asking a patient in pain to press 1 for scheduling.
What actually closes the gap is an agent that answers on the first ring at 9pm on a Sunday, sees real availability in your practice management system, and writes a confirmed appointment into it. That last clause is the hard one, and it is where the practice management system you already own determines what is possible.
Which practice management systems we have actually shipped against.
Most vendors publish a logo wall. This is the honest version: what runs in production today, how it connects, and what we would need to assess first.
| System | How it connects | Capability | Status |
|---|---|---|---|
| Open Dental | Official API, direct integration | Availability lookup, booking, modification, cancellation | In production |
| Curve | Official API, direct integration | Availability lookup, booking, modification, cancellation | In production |
| Dentrix | Windows-service gateway inside the practice LAN — no public API exists | Availability lookup and atomic booking: check, hold, confirm, or roll back | In production |
| Eaglesoft | Windows-service gateway inside the practice LAN — no public API exists | Availability lookup and atomic booking: check, hold, confirm, or roll back | In production |
| Curve Hero, Practice-Web, others | Assessed during discovery — API, gateway, or HL7/FHIR depending on what the system exposes | Scoped once we have seen the interface | Assessed first |
| No usable interface at all | Shared calendar your staff syncs | Booking capture without writeback into the PMS | Fallback |
The four systems marked in production are live in the CleverAnswerAI deployment across 20+ offices in the US and Canada. We list the rest as assessed first on purpose. We have not shipped against them, and a compatibility claim we cannot evidence is worth nothing to you at the point it fails.
Dentrix and Eaglesoft have no public API. Here is what we build instead.
Open Dental and Curve expose APIs, so integration is ordinary engineering work. Dentrix and Eaglesoft do not, and that is where most dental voice-AI projects stop at "the AI will take a message and your team enters it manually" — which is not booking, it is transcription with extra steps.
Our answer is a Windows-service gateway that runs inside each practice's own LAN. It handles two operations: availability lookup and atomic booking. Atomic matters here. The gateway does not fire a write and hope. It checks the slot, holds it, then confirms or rolls back, so two callers arriving in the same second cannot both be given the 10:30. Budget roughly two weeks of build per practice management system; that is real work, and pretending otherwise is how these projects slip.
Because the gateway lives inside the practice network, it also keeps the integration inside the covered boundary. Nothing about your schedule is mirrored to a third-party cloud in order to make booking work. If you want the full architecture — every vendor, what each one sees, and what an auditor gets — that is documented on our HIPAA-compliant AI receptionist page.
Generic speech models mishear dentistry. A 600-term glossary cut word errors by 38%.
A general-purpose speech model has no particular reason to know what occlusal means, or that SRP is scaling and root planing, or that Aetna PPO is one insurance product rather than three words. On a dental phone line those terms arrive constantly, over a compressed phone codec, frequently from a caller with a hand near their mouth.
For the dental network we built a 600-term custom vocabulary model covering procedures, insurance carriers such as Delta, MetLife and Aetna PPO, medications, and the abbreviations front-desk staff actually use. Measured on live-traffic calls, it cut word error rate by 38% against the generic model. That number is not a benchmark curiosity — a misheard carrier name or procedure is the difference between a booked appointment and a call that gets transferred.
Voice selection got the same treatment. A blind panel of 60 patients compared candidate voices and chose ElevenLabs over Azure Neural on warmth and perceived professionalism. We ran that as a panel rather than picking the voice the engineering team liked, because the person forming an impression of your practice is the patient, not us.
Six things the agent does. One category it always hands off.
Schedule, reschedule, cancel
Real availability read from your PMS, then an atomic write back into it. The agent calls six named tools and nothing else: check_availability, book_appointment, modify, cancel, collect_intake, transfer_to_human.
Intake before the first visit
Collects what the front desk needs from a new patient and writes a structured note, so the call arrives at your team already summarised rather than as a voicemail to replay.
Clinical questions always transfer
Whether tooth pain is an emergency, or whether to take more ibuprofen, is never answered by the agent. It refuses and warm-transfers to a human on-call. The boundary is explicit and deliberate.
20+ offices. Roughly 160K patient records. Zero HIPAA incidents.
Rolled out in tranches of five offices, with shared templates and per-location script variance.
A multi-location dental group with roughly 160,000 active patient records had trialled three commercial voice-AI platforms; all three failed HIPAA review. We built self-hosted LiveKit inside their VPC, negotiated BAAs with every vendor in the call path before any PHI moved, built the dental vocabulary model, and shipped in-LAN gateways for Dentrix and Eaglesoft alongside direct API integration for Open Dental and Curve. Fourteen weeks from HIPAA architecture document to 20+ live offices.
The 28% figure is measured across the first 90 days after deployment against the trailing 90-day baseline, on unchanged marketing spend — so it reflects calls captured rather than leads bought. The practice manager's summary was blunter than ours: the booking increase paid for the deployment inside the first quarter, and the unexpected part was that the front desk stopped being shouted at by patients who could not get through.
Single practice to multi-location group.
Fixed project pricing. Per-call run-rate is modelled during discovery so the monthly number is known before you commit.
| Deployment | Scope | Price |
|---|---|---|
| Single-office AI receptionist | Inbound voice, one PMS integration, scripted flows, shadow mode, staff handoff | $8,000–$14,000 |
| Multi-location rollout | Multi-office, shared templates, per-location config, analytics dashboard | $18,000–$35,000 |
| Recall & reactivation agent | Outbound voice and WhatsApp, treatment-aware cadence, PMS writeback, reporting | $10,000–$22,000 |
| Monthly retainer | Ops, script tuning, new intents, reporting, PMS change management | $2,500–$8,000/mo |
A single office goes live in 4–6 weeks including scripting, PMS integration, shadow mode and staff handoff. Multi-location rollouts add 1–2 weeks per practice management variant, because each closed PMS needs its own gateway built and tested. Engagements start with a discovery workshop at $1,500–$3,000, credited against the project. Full tiers are on the pricing page.
What dental practices ask on the first call.
Does it work with Dentrix, which has no public API?
Which dental practice management systems are already in production?
Will it understand dental terminology and insurance names?
Will it replace my front desk?
What does it cost per call?
How long does a dental deployment take?
Who wrote this, and why it matters when the decision is a compliance decision.
This page was written by Viktor Andriichuk, founder and lead architect at DataFlux Software. The PMS matrix above reflects integrations he built and runs in production, not a vendor feature list — which is why it names the systems where a gateway is required and the cases where we would assess before promising anything.
DataFlux Software has shipped 50+ production systems since 2018, and works on AI voice and agent infrastructure exclusively since 2023. Contracting entity: Gazolin Production SRL (Romania, EU).
The compatibility matrix, on one page.
The question to ask any vendor is not “do you integrate with Dentrix?” but “does the booking write back, and what happens when the write fails halfway?” This sheet is built to make that conversation short.
Tell us which PMS you run.
Twenty minutes, no pitch deck. Bring your practice management system, your monthly call volume, and your current missed-call rate if you have it. If an AI receptionist is the wrong answer for your practice, we will say so on the call.