Solution

Outbound Voice AI

An AI caller that reaches patients who need to schedule appointments—but only when your scheduling queue is quiet. Real conversations, not robocalls. Queue-safe by design, so inbound callers always come first.

OutboundVoiceAI

What it is

An outbound voice calling orchestrator purpose-built for healthcare scheduling. The system monitors your phone queue in real time and places AI-powered calls to patients who need appointments—only during quiet periods when agents are available. If the queue gets busy, outbound calling pauses instantly. Inbound patients never wait longer because of an outbound campaign.

Conversations are powered by OpenAI's Realtime API—not pre-recorded scripts. The AI adapts to what the patient says, answers operational questions about locations and prep instructions, and transfers to a human scheduler when the patient is ready to book.

The problem it solves

Patients abandon phone queues. Referrals sit unscheduled for days. Your scheduling team is too busy handling inbound calls to make outbound ones. And every day an appointment goes unbooked, you lose revenue and the patient loses time.

Traditional solutions don't work: hiring more staff is expensive, robocalls get ignored, and manual outreach campaigns are inconsistent. You need something that calls patients like a human, knows when to call, and never competes with real patients trying to reach you.

How it works

  1. Queue monitoring — Polls your phone system (FreePBX/Asterisk) every 10 seconds. Tracks agents available, calls waiting, and longest wait time.
  2. Gating check — Before placing any call, verifies agents are free, queue is empty, and conditions have been stable (no flapping). If anything fails, calling pauses immediately.
  3. Patient prioritization — Picks the next patient using a 4-tier priority system based on call history, due dates, and previous attempts.
  4. AI conversation — Places the call via Twilio. OpenAI Realtime handles the conversation: greets by name, qualifies availability, answers questions, and manages the flow.
  5. Resolution — Transfers to a human scheduler, leaves a voicemail with SMS follow-up, or flags the number for review—depending on the outcome.

Queue-safe by design

This isn't a dialer that floods your lines. The entire system is built around one principle: inbound patients come first, always.

Fail-closed architecture

If queue monitoring fails for any reason, outbound calling is disabled instantly. No guessing, no fallback—just stop.

Hysteresis gating

Queue must be quiet across multiple consecutive polls before calling resumes. Prevents flapping when volume fluctuates.

Pre-transfer re-check

Before connecting a patient to a human, the system re-verifies queue state. If it got busy during the conversation, the patient is offered a callback instead.

Single concurrent call

Only one outbound AI call at a time, globally. Your queue capacity is never consumed by parallel outbound sessions.

Smart patient prioritization

Not all patients are equal. The system uses a 4-tier priority system to decide who gets called first:

P1
Highest

Abandoned the queue and never received an AI callback. They tried to reach you and couldn’t.

P2
High

Abandoned the queue but already received one AI call. Still waiting to schedule.

P3
Medium

Called before but never got an AI follow-up. Engaged patients who need a nudge.

P4
Standard

Never called in at all. Referrals and new orders that need first contact.

Within each tier, patients are sorted by due date urgency, order age, and fewest prior attempts. The system respects business hours, holiday calendars, and configurable retry limits (default: 3 attempts, 6–8 hours apart).

Five call outcomes

Every call resolves to one of five outcomes, each with its own automated follow-up:

TransferredBest outcome

Patient is ready to schedule. AI re-checks queue, then connects them to a human scheduler in their preferred language.

Not availableRetry later

Patient answers but can’t talk now. AI offers a callback and sends an SMS with the scheduling number.

No answer / voicemailRetry later

AI leaves a HIPAA-safe voicemail (no PHI) and sends an SMS with callback details.

Wrong numberFlag for review

AI detects the number doesn’t belong to the patient. Sends an alert to the scheduling team for manual review.

Disconnected / invalidFlag for review

Carrier error detected. Number flagged and the scheduling team is notified via email.

Natural voice conversations

This isn't a phone tree with numbered options. The AI has a real conversation—powered by OpenAI's Realtime API—that adapts to what the patient says.

  • Greets patients by first name with facility-specific context
  • Answers operational questions: hours, locations, what to bring, prep instructions
  • Strictly avoids medical topics—no diagnoses, results, or medical advice
  • Handles interruptions, accents, and natural speech patterns
  • Multi-language support with automatic routing to language-specific queues

Real-time dashboard

  • Live queue status: agents available, calls waiting, outbound-allowed flag
  • Patient queue with priority tiers and next-up preview
  • Active call monitoring with live transcripts
  • System enable/disable toggle and business hours config

Full auditability

  • Searchable call history with complete transcripts
  • Dispatcher decision logs: why each call was or wasn't placed
  • Queue state snapshots for historical analysis
  • HIPAA-safe voicemails and SMS—no PHI in outbound messages

Built-in safeguards

Fail-closed on errors

Any monitoring failure immediately halts outbound calling

Phone number allowlist

Only pre-approved numbers can be called in production

Explicit enable flag

Live calling requires manual activation in settings

No PHI in voicemails

Outbound messages never contain protected health information

Retry limits

Max 3 attempts per patient with enforced cool-down periods

Business hours only

Calls respect configurable schedules and holiday calendars

Typical use cases

  • Medical imaging centers (MRI, CT, ultrasound scheduling)
  • Patients who abandoned the scheduling queue
  • Referrals and new orders that haven't been scheduled yet
  • Multi-language patient populations needing native-language outreach
  • Dental, physical therapy, and telemedicine scheduling
  • Any practice with high no-show or no-schedule rates
Case Study

See it in action

Learn how a California medical imaging chain combined our Support Agent, Email Automation, and Outbound Voice AI to cut support workload by 62% and boost patient satisfaction by 41%.