AI receptionist Planned

The opt-in flow: caller hears "press 1 to talk to our AI". If they press 1, AI handles the entire interaction. Appointment booking, order lookup, FAQ, after-hours messages. Pre-built vertical templates for restaurants, salons, vacation rentals, home services, medical, automotive.

Why opt-in

Mandatory AI receptionists are a brand liability. Customers feel trapped. The press-1 flow respects their preference: if they want a human, they get one (zero added friction); if they want AI, they get a faster path.

The flow

// Caller hears (cloned brand voice, configurable):
"You've reached Acme Auto Body. Press 1 to talk to our AI assistant
or stay on the line for a human."

// If they press 1:
// AI introduces itself, asks what they need, calls tools as needed,
// completes the request or hands off to a human.

1. Call hits the DID.
2. Cloned brand voice plays the press-1 prompt.
3. If 1, AI engages with the caller in conversation.
4. AI uses retrieval (KB) and tools (custom-defined per template) to handle the request.
5. If the request needs a human (tool failed, caller frustrated, explicitly asked), AI bridges to the queue.
6. Transcript and summary are written into the unified inbox like any other call.

Pre-built templates

Each template is a curated set of tools, prompt instructions, and escalation triggers. You can use them as-is or fork them in the dashboard.

• Restaurants. Reservations, hours, location, takeout.
• Salons. Appointment booking and rescheduling.
• Vacation rentals. Availability, booking, check-in instructions, lockbox codes.
• Home services. Quote requests, service-area check, dispatch.
• Medical. Appointment booking, prescription refill triage, after-hours nurse line handoff.
• Automotive. Service booking, parts inquiry, towing dispatch.

Configuration

Each DID can be configured independently. A typical body for restaurants:

// POST /api/v1/ai-receptionist/configure
{
  "phoneNumberId": "uuid",
  "enabled": true,
  "templateId": "restaurant_reservations",
  "afterHoursOnly": false,
  "voiceProfileId": "uuid",
  "tools": ["check_availability", "book_reservation", "modify_reservation"],
  "escalationTriggers": [
    "caller_frustrated",
    "request_to_speak_with_human",
    "tool_failed_twice"
  ]
}

Tools

Templates ship with sensible tool sets but you can add custom tools that call your systems. See custom tools for the function-calling protocol. Common patterns:

• Read-only lookups (order status, appointment list) usually run without confirmation.
• Write actions (book, refund, reschedule) prompt the caller for verbal confirmation, then execute.
• Destructive actions (cancel a paid order, charge a card) always escalate to a human.

After-hours mode

Set afterHoursOnly: true to keep the receptionist disabled during business hours (humans take the calls) and only engage after hours. This is the most common configuration. Business-hour schedules come from your workspace business hours.

Voicemail fallback

If the caller does not press 1 and no human is available, the call goes to voicemail. Transcription is automatic via our speech-to-text pipeline; a voicemail row is created with the recording URL and transcript. See voicemails API.

Compliance and disclosure

The greeting always identifies the assistant as AI. We never let it pretend to be human. For US calls we honour TCPA (Telephone Consumer Protection Act) requirements and surface call-recording disclosure on a per-jurisdiction basis. See TCPA compliance concept.

Common pitfalls

• Too many tools. AI degrades when the toolbox is huge. Pick a template and add tools sparingly.
• No escalation triggers. Always include "request to speak with human". Some callers want a human and pressed 1 by mistake.
• Recording without disclosure. Some US states require two-party consent. Use our default disclosure copy or replace it with your legal team's version.