Healthcare

HIPAA-grade voice agents for a multi-state clinic group

We replaced a 38-person inbound desk with sub-second voice agents that handle 78% of patient calls without escalation — and made the remaining 22% measurably better.

Calls auto-resolved
78%
Annual savings
$3.4M
Patient CSAT (10)
9.2
p95 turn latency
210ms
The challenge

Inbound volume had grown 31% YoY against a frozen support headcount. The CIO's mandate was unambiguous — automation that nursing leadership and counsel both signed off on, or no automation at all.

Our approach

We started with a 5-day discovery sprint: stakeholder interviews, a sample of 1,200 historical calls hand-labeled into 17 intents, and a HIPAA risk model. From there, we built a streaming voice pipeline (Twilio → Deepgram → policy gate → GPT-4o-mini routing → ElevenLabs) with PII redaction at every hop and a per-call lineage record auditors could query.

The outcome

After 11 weeks, the system handled 78% of inbound volume autonomously with a 9.2 CSAT — higher than the human baseline of 8.4. Operations re-deployed 21 of the original 38 desk staff to clinical care coordination roles.

We expected a deflection bot. We got an agent our nursing leadership trusts to talk to patients before they reach a person.

Dr. Renée Allard · CIO, Smile Dental
Project details
Client
Smile Dental
Industry
Healthcare
Duration
11 weeks
Region
Multi-state, US
Tech stack
Twilio VoiceDeepgramGPT-4o-miniElevenLabsPostgresTemporalDatadog
Services delivered
Voice AI DevelopmentCompliance ArchitectureEmbedded AI Team

Similar challenge?

Let's scope your project

Book a call and tell us what you're trying to ship. A senior engineer will respond within one business day.