Voice AI for appointment scheduling in healthcare

Integration touchpoints

• EHR read for patient demographics, problem list, and slot rules (provider, location, visit type)
• Practice-management write for the booking itself, with full reason coding
• Eligibility / clearinghouse read so the AI surfaces coverage status during scheduling
• Reminder cadence owned by the booking interaction, not by a separate product

Regulatory hooks

• HIPAA — BAA with every sub-processor in the call path, minimum-necessary disclosure on demographics
• 21st Century Cures Act — information-blocking rules apply if AI gates access to records
• State medical-board rules — scheduling is fine, clinical triage requires licensed oversight
• ADA / Section 508 — the AI front door has to fail open to a human, not to a dead end

What good looks like

Patient identifies themselves, the AI confirms demographics, surfaces eligibility status, negotiates a slot against real practice-management availability, books with reason code, and triggers the reminder cadence. Anything that looks like a clinical question (symptom assessment, urgency, medication interaction) is routed to a licensed human with full context.

Watch-outs

• Symptom triage drift. The AI must refuse clinical judgement firmly and route, not improvise.
• Booking against a flat day grid. Sells slots clinics cannot keep; the rescheduling cost erases the containment gain.
• Sub-processor BAAs missing on the model provider. PHI exposure inside the LLM context window without a BAA is a breach event.
• Skipping accessibility testing. TTY and relay-service users are a regulated subset.

Frequently asked