AI Voice Agent: Inbound Receptionist (Retell + n8n)

Author / channel: Zubair Trabzada Format: video Source: Original Published: 2025-06-28

Summary

End-to-end no-code build of an inbound receptionist voice agent for a dental office: Retell AI handles voice + persona + tool calling (cal.com calendar check/book, transfer, end-call), n8n handles post-call CRM updates (Google Sheets / HubSpot) via a webhook + structured output parser. Frames the business case with industry stats: dental clinics miss 27% of inbound calls, auto shops 39%, banks 51% — every missed call is lost revenue. Single-prompt agent setup with GPT-4.1 Mini (~$0.094/min). Generalizes to any service-based vertical.

Key Points

  • Business case — Invoca data: dental 37% missed calls, auto dealerships 37%, auto service 39%, banks 51%. Voice agents are a solved-problem revenue play.
  • Stack splitRetell AI = voice front end + persona + function calling + telephony. n8n = post-call automation backbone (webhook → AI agent → CRM update).
  • Retell agent type — single-prompt (vs multi-prompt or conversation flow). Easiest to build, flexible enough for receptionist work.
  • Voice agent functionscheck_calendar_availability, book_appointment (both via cal.com API + event_type_id), transfer_call (cold/warm with target number), end_call.
  • Cost — GPT-4.1 Mini “Fast” voice ~$0.094/min. Higher-quality voices add cost.
  • n8n side — webhook (POST, prod URL active, test URL for builds) → AI agent node with structured output parser (JSON schema for first/last name, phone, email, insurance, call summary) → Google Sheets row update / HubSpot upsert / email / Slack.
  • Structured output trick — enable “require specific output format” on the n8n AI agent node, attach a structured output parser, paste your JSON schema. This is how you control LLM output shape reliably.
  • Knowledge base — Retell supports a separate knowledge base attachment, but the demo embeds business info inline in the persona prompt for simplicity.

Connected Pages