Zero missed calls and 31% fewer no-shows for a dental group
A multi-location dental group missed every call after close and couldn't keep up during the day. We deployed an inbound voice agent integrated with their practice-management system that books, reschedules, and confirms in English and Spanish, around the clock, escalating only clinical questions to staff.
- 0
- Missed calls
- 31%
- Fewer no-shows
- 68%
- Booked fully by AI
What they came with
Calls arrived at lunch, after hours and while the front desk was already with a patient, and a call a practice does not answer is usually a booking made somewhere else. Voicemail did not close the gap, because returning it competed with the same queue at the counter. Appointments that were booked still left chairs empty when people forgot, and reminders were a job the desk fitted in when there was time.
What the engagement covered
- Real-time calendar booking with no double-bookings
- Bilingual conversation with automatic language detection
- Confirmations and reminders that drove no-shows down
Technical detail
Streaming audio, not recordings
Twilio media streams push call audio to Deepgram as it arrives, and partial transcripts start the model turn before the caller has finished the sentence. Playback is cut as soon as speech is detected, so the caller can interrupt the agent.
Booking holds the slot first
The agent reads live availability from the practice management system and holds the slot before confirming it aloud, so two callers cannot be given the same time. If the write fails the call transfers to the desk rather than confirming an appointment that does not exist.
Reminders scheduled off the appointment
The confirmation and reminder sequence is scheduled from the appointment record and cancels itself when the appointment moves or is cancelled. Each job is idempotent, so a retry does not ring a patient twice.
Escalation carries the transcript
Anything outside the agent's scope transfers to the desk with the call summary and transcript attached. Whoever picks up starts from what was already said instead of from the beginning.
The stack
Telephony
Speech and AI
Backend
Data
- Practice
- Inbound Voice Agent
- Sector
- Healthcare
- Shape
- Client engagement
- Stack
- OpenAI Realtime, Twilio, Deepgram
Something like this to build?
Tell us what runs today and where it hurts. An engineer reads it and replies.