An outbound voice agent that holds a real conversation.

End-to-end STT → LLM → TTS pipeline that books callbacks, handles objections, and writes notes back to Salesforce — with sub-1-second p95 latency.

1,400

staff hours saved per month

+28%

conversion lift on policy renewals

<1s

p95 latency, end-to-end

Insurance · EU

Insurance — EU

Client

Insurance — EU

Insurance · EU

Headline metric

1,400

staff hours saved per month

Deliverables

shipped to production

Stack

5+ tools

across the build

Challenge

An EU insurer was paying a 60-person callback team to handle policy renewals and missed-claim follow-ups. They had a 22% pickup rate, 8% conversion, and a 9-week training ramp per new hire. The team couldn't scale into new markets.

Approach

Built an end-to-end STT (Deepgram) → Claude → TTS (ElevenLabs) pipeline on LiveKit with sub-1s p95 latency. The agent handles objections from a structured playbook, books callbacks into Calendly, and writes notes back to Salesforce. SOC 2-aligned audit logs on every call.

Outcome

1,400 staff hours saved per month, 28% conversion lift on renewals, and the agent now runs 18 hours a day across three languages. The human team moved up to complex claims work — and morale went with it.

“

1,400 staff hours saved per month, 28% conversion lift on renewals, and the agent now runs 18 hours a day across three languages. The team owned this end-to-end.

Insurance — EU

Insurance · EU

1,400

staff hours saved per month

+28%

conversion lift on policy renewals

<1s

p95 latency, end-to-end

Deliverables