Solo dental practices miss 30-35% of incoming calls. I've been working on an AI system that fixes this — here's the technical breakdown.
The Problem Space
A solo dentist gets ~40 calls/day. They're chairside most of the time. No budget for a second receptionist. Result: 12-14 missed calls daily, each worth €800-€1,200 in lifetime patient value.
Architecture Overview
The system needs to:
- Answer inbound calls with sub-200ms latency
- Understand natural speech (including regional accents)
- Book appointments in real-time against a calendar API
- Triage emergencies vs routine queries
- Handle concurrent calls (no busy signals)
Key Technical Decisions
Speech-to-Text: We need real-time streaming ASR with low latency. Whisper-based models fine-tuned on medical/dental terminology work well here, but you need edge inference for the latency budget.
Conversation Engine: The tricky part isn't understanding what the caller says — it's maintaining context. "I need a cleaning, but also my kid needs braces, and can we do both on Thursday after 3?" That's three intents, two patients, and a scheduling constraint in one sentence.
PMS Integration: Dental practice management systems (Dentally, SOE, etc.) have varying API quality. Some are REST, some are SOAP nightmares from 2008. Build a robust adapter layer.
Emergency Detection: False negatives here are unacceptable. If someone calls saying "I fell and my tooth is hanging out," that needs to hit the dentist's mobile in under 30 seconds. We use a conservative classifier that over-escalates rather than under-escalates.
Results
For practices using this system:
- Answer rate: 65% → 99%+
- After-hours bookings: 0 → 15-20% of total
- Patient satisfaction: Higher (no hold music, no phone trees)
- Cost: <€200/month vs €35,000/year for a human
What I Learned
- Accent handling is everything in a local market. A system trained on American English is useless in Cork.
- Dental terminology is its own language. "I need a composite" vs "I need a filling" vs "I need the white one" are the same request.
- Trust takes time. Dentists are (rightfully) cautious about automating patient interactions. Free trials with full transparency matter.
If you're building in the voice AI space, I'd love to compare notes. What latency targets are you hitting for real-time conversation?
VoiceFleet — AI receptionists for dental and healthcare practices.
Top comments (0)