Appointment no-shows cost millions. Reminder calls work — but staff time doesn't scale, and customers screen unknown numbers.
The AI approach
Conversational TTS + STT. Voice agent places the call, confirms or reschedules in natural conversation; hands off to human only when confused or hostile.
The outcome
~−24% no-shows
Reach rate up; no-show rate down ~24%. SDR time freed for genuine sales calls. Patients/customers feel reminded, not nagged.
Try itInput → Process → Output
Input — appointment list
Khun Anan S.
cardio · Wed 14:00 · Dr. P'Boon · 089-***-1422
Khun Patcharee M.
dental · Wed 16:30 · Dr. P'Mai · 086-***-3344
Khun Vorawit T.
follow-up · Thu 09:00 · Dr. P'Boon · 081-***-5677
Khun Sirikit M.
physio · Fri 11:00 · Dr. P'Nan · 084-***-2231
4 calls due · 24h before appt
Process — AI pipeline
1Place outbound call (TTS)SpeakGenerative
2Listen + transcribe responseListenPredictive
3Apply intent + branching rulesRulesSymbolic
4Confirm / reschedule / handoffSpeakGenerative
Output — call outcome
Click Run demo to place the call and resolve confirm/reschedule.
Live call · Khun Anan resolved · 47s
ConversationAI: "Hi, this is the Bumrungrad reminder line for Khun Anan. You have a cardio check-up tomorrow at 2pm with Dr. Boon. Will you be coming?" Customer: "Yes I'll be there." AI: "Great — see you at 2pm. Just press 9 anytime to opt out of these reminders."
Today's outcomes 4 calls · 38s avg
Khun Anancardio · Wed 14:00Confirmed
Khun Patchareedental · Wed 16:30Rescheduled · Thu
Khun Vorawitfollow-up · Thu 09:00Confirmed
Khun Sirikitphysio · Fri 11:00Hand off to human
Why hand-off: customer asked about insurance coverage — outside agent's policy scope. Routed to front-desk callback.
Three AI types in this use case
SymbolicConversation tree (confirm / reschedule / opt-out / handoff); compliance disclaimers (TCPA / DNC); calling-window rules per timezone.
PredictiveSTT for utterances; intent classification; sentiment scoring for handoff-to-human signal.
GenerativeTTS in a friendly natural voice; LLM handles unexpected utterances ("can I bring my mom?") within policy scope.
The stack
Telephony · Twilio / Vapi / Retell
TTS · ElevenLabs / Azure Neural
STT · Deepgram Nova
LLM · Claude with policy prompt
When this works
Verticals where reminders move the needle (clinics, salons, services)
Calling-window compliance is automated by timezone
Opt-out path is one-step ("press 9 to stop")
Volume justifies per-call cost (~$0.10–0.30)
When it fails
Hostile / confused customer — needs handoff fast
Voice is too good — customer thinks it's a real person
Calling outside permitted hours violates TCPA
"I lost my ID" reasoned away when it should escalate