Shipping an Enterprise Voice AI Agent in 100 Days
summary
Finn Voice goes from blank page to a deployed enterprise phone agent in about 100 days, using real‑time APIs and a stack of STT–LLM–TTS + RAG + telephony while starting with an out‑of‑office wedge and a lightweight Playground for testing and monitoring. It hits about 56% average resolution rate on chat‑style KB queries, targets ~1s latency for simple answers with filler for longer ones, and claims costs of 3–20¢ per minute versus $7–$12 human calls.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 0:44About Intercom
- 1:11Intercom Fin
- 1:43Continuous Optimization
- 1:55Voice Channel
- 2:56Benefits of Voice
- 3:37How Intercom Fin came about
- 4:16Use case
- 6:24What we shipped first
- 6:53Fin Voice Playground
- 7:26Deploy Experience
- 7:43Monitoring
- 8:05Text Stack
- 9:40Conversation Design
- 12:26Integrations
- 14:53Pricing
- 16:00Final Thoughts
related talks
56mAI Agents That Don’t Suck: How to Build Ones That Actually Drive Business Value
33mDesigning Voice Agents for Real Conversations
17mAI Voice agents for the Business and Phone Channels
36mHow We Built a Voice Agent with Temporal
51mCan AI Face Your (Potential) Customers?
20m10 Things I Learned Building Voice Agents in 2025
25mEngineering voice agents: Latency, quality, and scale
08mGive Your Chat Agent a Voice