AI Voice Automation for HealthFirst Provider Network
We built an AI voice agent that handles 82% of incoming calls for a 14-clinic healthcare network — scheduling appointments, processing prescription refills, and triaging symptoms at $0.40 per interaction versus $7.20 for human agents.
The Challenge
HealthFirst Provider Network received 4,800 phone calls per day across its 14 clinics. Of those, 78% fell into three categories: appointment scheduling/rescheduling (42%), prescription refill requests (21%), and general inquiries about hours/locations/insurance acceptance (15%). Despite these being routine, predictable interactions, each one required a human receptionist to answer.
The network employed 28 full-time receptionists at an annual cost of $1.96M. During peak hours (8-10 AM and 3-5 PM), hold times averaged 4 minutes and 23 seconds. Patient satisfaction surveys showed phone accessibility as the number-one complaint — 34% of respondents rated it "poor" or "very poor." The clinics estimated that 12% of callers hung up before reaching a person, representing approximately 576 missed interactions daily.
Previous attempts at automation — an IVR tree installed in 2021 — had made things worse. Patients hated pressing buttons through nested menus, and the system couldn't handle the natural language of real requests. "I need to see Dr. Chen sometime next week, preferably morning" can't be processed by "Press 1 for appointments."
Our Solution
We replaced the IVR with a conversational AI voice agent powered by Claude for natural language understanding and ElevenLabs for natural-sounding speech synthesis. The agent handles the top 3 call types end-to-end: it checks provider availability in real-time via the practice management system (connected through HL7 FHIR), books/reschedules appointments, processes prescription refill requests by routing them to the correct pharmacy queue, and answers frequently asked questions from a dynamic knowledge base.
The voice UX was critical. We designed the agent to sound like a competent, unhurried human — not a robot. It confirms understanding by paraphrasing ("So you'd like to see Dr. Chen next Tuesday or Wednesday morning — let me check her schedule"), handles interruptions gracefully, and transfers to a human agent immediately when it detects medical urgency keywords or when the caller explicitly requests a person.
The system integrates with Twilio Voice for telephony, the EHR system through HL7 FHIR APIs for appointment and patient data, and a custom analytics dashboard that tracks containment rates, transfer reasons, call durations, and patient satisfaction scores collected via post-call SMS surveys.
Implementation Process
1
Call Analysis & Design(3 weeks)
Analyzed 10,000 recorded calls to map conversation patterns, designed 47 dialog flows, and created the voice persona with the clinical team's input.
2
AI Agent Development(6 weeks)
Built the Claude-powered NLU engine, ElevenLabs voice synthesis pipeline, Twilio telephony integration, and HL7 FHIR connectors to the EHR system.
3
Testing & Compliance(3 weeks)
Ran 2,000 test calls across all 47 dialog flows, HIPAA compliance review, penetration testing on the telephony system, and staff training.
4
Phased Rollout(4 weeks)
Launched at 2 clinics, monitored for 2 weeks, refined based on real call data, then expanded to all 14 clinics over 3 weeks.
Results
After 90 days across all 14 clinics, the AI voice agent had handled over 350,000 calls. The results exceeded the initial projections.
82%
Containment rate
82% of calls fully resolved by AI without human transfer
$0.40/call
AI cost per interaction
vs. $7.20 per human-handled call — 94% cost reduction
28 sec
Average handle time
Down from 3 min 42 sec with human agents for routine calls
0 sec
Hold time
AI answers every call instantly — no queues, no hold music
$1.2M
Projected annual savings
Receptionist staff reduced from 28 to 12 (reassigned to patient care roles)
Technology Stack
C
Claude API (Anthropic)
Natural language understanding and dialog management
E
ElevenLabs
Natural-sounding voice synthesis
T
Twilio Voice
Telephony infrastructure and call routing
P
Python 3.12
AI agent logic and HL7 FHIR integration
P
PostgreSQL 16
Call logs, analytics, and configuration
H
HL7 FHIR
Healthcare data interoperability with EHR systems
Ready to Start?
No commitments. Tell us what you need and we'll tell you how we'd solve it.
“Patients tell us they didn't realize they were talking to an AI until we mentioned it. That's the highest bar for voice technology — when people can't tell the difference, and more importantly, when they don't care because the problem got solved.”
— Chief Operations Officer, HealthFirst Provider Network
Ready to Transform Your Phone Operations?
If your team handles hundreds of routine calls daily, AI voice agents can resolve 70-85% of them at a fraction of the cost. Let's explore what's possible.
Free call analysis · We'll estimate your containment rate