Building AI-Powered Schedule Gap Prevention for Healthcare
I've been working on an interesting problem: automatically filling cancelled appointments in dental practices. The domain is healthcare, but the engineering challenge is broadly applicable to any booking system.
The Problem Space
When a patient cancels a dental appointment same-day, practices have a narrow window to fill the slot. Traditional approach: receptionist manually calls waitlist patients one by one. Fill rate: ~20-30%.
The constraints are interesting:
- Time pressure: Often < 2 hours to fill the slot
- Sequential bottleneck: Human can only call one person at a time
- Match complexity: Need to match treatment type, duration, insurance, and patient availability
- High cost of failure: Each empty hour ≈ €200-300 in lost revenue
The Architecture
The system monitors the practice management software for schedule changes via webhooks/polling. When a cancellation is detected:
- Filter the waitlist — match by treatment type, expected duration, insurance compatibility
- Score candidates — proximity to practice, historical reliability, urgency of their need
- Parallel outreach — send SMS + automated calls simultaneously to top candidates
- First-confirm-wins — race condition solved with simple locking: first confirmation books the slot, others get "slot filled" response
- Confirmation loop — update PMS, send patient confirmation, alert staff
The tricky part is the matching algorithm. A 30-minute hygiene cancellation can't be filled with a 90-minute crown prep. The AI needs to understand treatment type compatibility and time-fitting.
Learnings
- SMS beats calls for younger patients, calls win for 55+
- 15-minute buffer between notification and slot time dramatically improves show-up rate
- Over-notification kills trust — patients who get too many "we have an opening!" messages start ignoring them
- Fill rates jumped to 60-80% with this approach vs 20-30% manual
If you're building anything in the appointment/booking space, the parallel-outreach-with-locking pattern is worth considering.
This is part of the work we're doing at VoiceFleet — AI-powered phone handling for healthcare practices. Curious about the technical side? Happy to discuss in comments.
Top comments (0)