AI Voice Agent System

Outbound Voice Agent for Auto Garage

n8nRetell AIGoogle SheetsCal.com

Executive Summary

Automated outbound calling system that reactivates past customers with personalized offers. Agent "Jacob" handles end-to-end calls, from greeting to appointment booking, with zero manual dialing.

Workflow

n8n workflow: Trigger to Google Sheets to Retell API
Retell AI agent configuration and prompt

System Architecture

Components

  • Orchestrator: n8n workflow (trigger → fetch → API call)
  • Voice Engine: Retell AI LLM with custom agent persona
  • Data Layer: Google Sheets (customer CRM)
  • Scheduling: Cal.com API for availability + booking

Boundaries

  • n8n owns batch execution; Retell owns conversation state
  • Customer data stays in Sheets; call metadata returns via webhook
  • Booking confirmations trigger downstream email automation

Agent Design

Tool Calling

  • check_availability_cal — fetch open slots
  • book_appointment_cal — confirm booking
  • lookupCustomer — verify contact info
  • sendConfirmation — trigger SMS/email

Routing Logic

  • Intent: schedule → offer slots → book
  • Intent: not interested → polite close + deadline mention
  • Intent: speak to human → transfer_call

Guardrails

  • Offer validity date hardcoded (July 31, 2025)
  • No pricing discussion; transfer if asked
  • Max 2 rebuttals before graceful exit

Escalation

  • Angry/confused → immediate human transfer
  • Booking failure → log + flag for manual followup
  • Voicemail detection → skip + mark "no answer"

Reliability & Ops

Implemented

  • Retell API rate limiting (10 concurrent calls)
  • Row-level "called" flag prevents duplicate dials
  • Webhook retry on 5xx (3 attempts, exponential backoff)
  • Call status tracked in Sheets (pending/completed/failed)

Next Iteration

  • Circuit breaker for API outages
  • Dead-letter queue for failed bookings
  • Automatic rescheduling on no-answer

Observability & QA

Implemented

  • Full transcript stored per call (Retell dashboard)
  • n8n execution logs with input/output per node
  • Call duration + outcome metrics in Sheets

Next Iteration

  • LLM-as-judge for transcript quality scoring
  • Sentiment analysis on call recordings
  • Weekly conversion funnel dashboard

QA Checklist

  • Agent introduces itself correctly
  • Offer details accurate (date, service)
  • Slots offered match calendar availability
  • Booking confirmation sent successfully
  • Escalation triggers work as expected
  • Graceful handling of edge cases

Impact

100%

Reduction in manual dialing time

Dozens

Customers contacted per batch run

Real-time

Slot availability + instant booking

Full

Operational visibility via Sheets

Downloadable Artifacts

Case Study PDF

Full documentation with screenshots

n8n Workflow JSON

Import directly into n8n

Interested in a similar system?

I build AI voice agents and automation workflows for customer outreach, scheduling, and operations.

Get in TouchView More Projects