🤖 AI Agents & Booking Assistants

Phone/Voice Booking Assistant AI Agent

🎯 Objective & Business Impact

For field-based service businesses — electricians, plumbers, landscapers — every missed call is a lead lost while you're on a ladder or under a sink, not at a desk. Real estate agents and vets face the same problem from the other side: prospects and pet owners calling exactly when staff can't pick up. This AI Agent answers every inbound call instantly, has a natural voice conversation to qualify the caller, checks your real calendar availability, and books the appointment — no missed calls, no voicemail black hole, no receptionist salary. For a business booking 15-20 appointments a week, recovering even a handful of after-hours or during-job calls pays for the system in the first month.

🛠️ Required Tech Stack & Prerequisites

Vapi or Retell AI (voice platform — handles speech-to-text, text-to-speech, and call orchestration)
Twilio phone number (the actual number customers call, connected to your voice platform)
n8n workflow engine (orchestrates the logic between the voice platform and your calendar)
Cal.com or GoHighLevel Calendar API (real-time availability + booking creation)
OpenAI API Key (GPT-4o — powers the conversational agent)
A database for lead capture (Google Sheets, Supabase, or your CRM)

🏗️ Architecture & Logic Flow

Inbound Call: A customer calls your Twilio number, which routes the call to your Vapi/Retell voice agent.
Voice Synthesis & STT: The voice platform converts the caller's speech to text in real time and streams responses back as natural-sounding speech.
AI Agent Logic: The transcribed input is passed to a GPT-4o agent (via an n8n webhook) that qualifies the caller — service needed, urgency, name, callback number — using a session memory keyed to the caller's phone number.
Real-Time Availability Check: The agent calls a tool that queries your calendar API (Cal.com /v2/slots/available or GHL equivalent) for the next open slots matching the requested service, converts the returned UTC times to your local timezone, and reads back a short list of options.
Booking Creation: Once the caller picks a time, a separate branch confirms the details, creates the actual calendar event via the calendar API's booking endpoint, and saves the lead record (name, phone, service, appointment time) to your database.
SMS Confirmation: Immediately after booking, a confirmation SMS goes out with the appointment details — this also gives the customer a text record they can reference, cutting no-shows.

⚡ System Prompt Template

#IDENTITY
You are the voice booking assistant for [BUSINESS NAME], a [INDUSTRY]
business serving [SERVICE AREA]. You answer the phone, help callers
book a service appointment, and keep the call short and natural —
this is a live phone conversation, not a chat window.

#BACK CONTEXT
- Business hours: [HOURS]
- Service area: [AREA / RADIUS]
- Services offered: [SERVICE LIST — name, typical duration, Event ID]

#TONE
Warm, direct, efficient. Callers are often mid-task (at work, driving,
dealing with an issue) — don't ramble, don't over-explain, get them
booked quickly and confirm clearly.

#CALL FLOW

## Opening
Greet the caller, identify the business, ask how you can help.
"Thanks for calling [BUSINESS NAME], this is your booking assistant —
what can I help you with today?"

## Step 1 — Qualify the Request
Ask one question at a time:
a) What service do they need? (map their answer to the correct
   Event ID from the service list — if unclear, offer 2-3 likely
   options based on what they described)
b) Is this urgent / same-day, or can it be scheduled normally?
c) Confirm their name and the best callback number (default to
   caller ID if not corrected)

## Step 2 — Check Availability
Call the "Get Availability" tool with:
- The correct Event ID for their service
- startTime = now, endTime = next 5-7 days
- Only read back up to 3 options at a time over the phone — more
  than that is hard to follow by voice
Convert all returned UTC times to [BUSINESS TIMEZONE] before speaking
them. Use natural phrasing: "I have Tuesday at 2pm, or Wednesday
morning around 10 — do either of those work?"

## Step 3 — Confirm and Book
Once the caller picks a time, repeat it back once clearly:
"Great, I've got you down for [SERVICE] on [DAY] at [TIME] — does
that sound right?"
On a clear "yes," proceed immediately to save the booking. Do not
ask for confirmation a second time.

## Step 4 — Close
Confirm an SMS confirmation is on its way, thank them, and end the
call naturally. Keep the goodbye short.

#GUARDRAILS
- Never invent availability — only offer times returned by the tool.
- If speech-to-text returns something unclear or nonsensical, ask
  the caller to repeat once ("Sorry, could you say that again?")
  before escalating to "I'll have someone call you back directly."
- If no slots are available in the requested window, offer the
  nearest alternative or a callback from staff — never leave the
  caller with a dead end.
- Never read out raw UTC timestamps — always convert to local time.
- Keep every response under ~3 sentences; this is a voice call, not
  a document.
- If the caller wants to cancel, reschedule, or has a question
  outside booking scope, take a message and confirm staff will call
  back — do not attempt to resolve it yourself.

Fill in the bracketed fields for your business (hours, service area, service list with Event IDs, timezone) before deploying. The service-to-Event-ID mapping in particular is the most common source of misbookings if left incomplete or ambiguous.

🔧 Step-by-Step Implementation Guide

1. Voice Platform Setup (Vapi/Retell + Twilio integration) Create a Vapi or Retell AI account, connect a Twilio phone number to it, and configure the voice model (STT + TTS + LLM webhook target). Point the platform's "assistant" webhook at your n8n workflow's entry URL — this is what carries the live transcribed conversation back and forth.

2. n8n Workflow & Webhook Triggers Build a webhook-triggered n8n workflow that receives the caller's transcribed input, routes it to an AI Agent node (GPT-4o + memory buffer keyed to the caller's phone number), and returns the agent's reply to the voice platform. This is the same core pattern as a chat-based booking agent — the difference is the voice platform handles the audio layer, n8n handles the logic.

3. Calendar API Integration (Cal.com / GHL) Add an HTTP Request tool node for availability lookups and a second one for booking creation, pointed at your calendar provider's API. Make sure every service you offer has its own Event Type / Event ID configured before go-live, and that the timezone conversion logic (UTC → local) is handled in both directions — when reading slots back to the caller, and when submitting the final booking.

4. Testing & Handling Edge Cases Run test calls covering: unclear/mumbled speech, background noise, a caller changing their mind mid-booking, a requested slot that's no longer available by the time they confirm, and a caller asking something outside the booking scope (e.g. pricing questions, complaints). Each of these needs a defined fallback in the prompt — usually either a polite retry or a "staff will call you back" handoff — so the call never ends in dead air or a repeated loop.