CASE STUDY · RESTAURANTS & HOSPITALITY

A Voice Agent That Takes Reservations While the Kitchen Is Slammed

A restaurant was missing reservation calls during service. We built a voice agent that answers, checks real table availability, takes the booking with party size and special requests, and logs every call.

Google Workspace · Automated Booking Confirmation Email
Restaurant Reservation Confirmation Email
🍽️ Industry
Restaurants & Hospitality
🎙️ Service
📋 Handles
Bookings, Times, Queries, Transfers
⚙️ Platforms
VAPI & Make.com
🧠 AI Models
GPT-4o, Deepgram, ElevenLabs
📊 Logging
Google Sheets (Full Call Record)
🟢 Status
Live in Production

The Phone Rings Hardest When Nobody Can Answer It

A restaurant’s busiest phone hours are its busiest floor hours.

Someone calls at 7pm on a Friday to book a table for six, and every person who could answer is carrying plates. So the call rings out. The caller does not leave a voicemail about a dinner reservation — they book somewhere else, that night, for six people.

The calls that do get answered are handled mid-service, at a podium, by someone who is also seating a walk-in. Party size gets misheard. The special request does not make it to the book. Nothing is recorded.

What We Built

Four foundational pillars engineered for seamless table reservations:

📞

An Agent That Answers Every Call

Named, warm, and conversational. It handles reservations, answers questions about the restaurant, and hands off to a person when the call needs one.

📖

A Knowledge Base It Answers From

Hours, menu, location, and house policies are loaded in, so questions get answered accurately rather than deflected.

📅

Live Availability Checking

The agent queries the real calendar during the conversation and offers times that are genuinely open.

📊

A Record of Every Call

Caller details, party size, special requests, written summary, recording, duration, and cost written to a tracking sheet automatically.

VAPI · Assistant Knowledge Base & System Prompt Configuration
VAPI Restaurant Agent Configuration
ANNOTATION: Knowledge base loaded with hours, menu & policies ANNOTATION: System prompt configured for conversational table booking

Checking Availability Is Harder Than Taking a Booking

Writing a booking into a calendar is simple. Working out what is actually free, in real time, and describing it to someone out loud is not.

A calendar returns availability as raw blocks of busy and free time. A caller cannot be read a data structure. They need to hear something like "we have 7:15 or 8:45 tonight."

That translation is what this part of the system does. It queries the calendar for free and busy periods, steps through the results one at a time, and converts them into a short set of times the agent can offer conversationally — then routes down separate paths depending on whether anything is available. It is the least visible part of the build and the part that makes the conversation feel normal.

Make.com · CheckAvailability Real-Time Calendar Pipeline
Make.com CheckAvailability Scenario
1. Agent asks what is available 2. Checks real free and busy times 3. Steps through each opening 4. Turns raw times into something speakable

What Happens on a Call

Seven automated stages from the first ring to confirmed table reservation:

01 STEP 1

The Agent Answers

Greets the caller by the restaurant’s name and asks how it can help.

02 STEP 2

Works Out Intent

A reservation, a question about hours or the menu, or something that needs a person.

03 STEP 3

Answers from Knowledge Base

Hours, location, menu, and policies answered instantly, without guessing.

04 STEP 4

Takes Reservation Details

Captures date, time, party size, and any special requests the caller mentions.

05 STEP 5

Checks Real Availability

Queries the live calendar and offers table openings that actually exist.

06 STEP 6

Confirms and Books

Repeats the details back, writes the reservation, and sends confirmation.

07 STEP 7

Logs the Call

Caller details, outcome, summary, recording, duration, and cost written to sheet.

HUMAN HANDOFF

Warm Transfer Ready

If the caller asks for a person at any point, the call transfers and is logged.

The Reservation It Took

A table for six, booked during the call, written into the restaurant's calendar with the party size and the caller’s special request attached, with the confirmation dispatched automatically and a reminder set.

Confirmation Email Dispatched
Confirmation Email Evidence
✓ Sent automatically upon call completion
Google Calendar Event Created
Google Calendar Event Evidence
Party size & special request captured Created by the system, not a person

⚡ Nobody on the floor touched any part of this booking.

Every Call, on the Record

This is the part most people do not expect. Every call the agent takes leaves a complete, structured record.

Every call leaves a record: who called, their number and email, whether the call was transferred to a person, whether a reservation was made and for when, how many guests, what they asked for, a written summary, a link to the recording, how long the call ran, and what it cost to handle.

A phone answered by a person leaves none of that behind. You get a name in a book, if you are lucky, and whatever the host remembers. This logging sheet is also what makes reporting possible later — call volume by hour, which nights run hottest, and caller intent patterns.

Google Sheets · Full Inbound Call Log & Telephony Audit Trail
Restaurant Call Tracking Sheet
Every call recorded & summarized Transfers tracked with reasons Cost & duration logged per call Foundation for custom dashboard analytics

How It's Wired

A resilient production pipeline tuned for low latency and conversational coherence:

Speech & Understanding
Deepgram transcribes in real-time, while GPT-4o runs the conversation against a defined identity, a set of abilities, and the restaurant's own knowledge base. ElevenLabs Turbo v2 handles voice synthesis, chosen for ultra-low latency.
Knowledge Base
Hours, menu, location, and house policies are loaded as modular documents the agent queries, rather than being hardcoded into the prompt. That means menu items or holiday hours can be updated without rewriting the agent's behavior.
Availability Checking
A webhook fires into a Make.com scenario that queries Google Calendar for free and busy periods, iterates over the results, and converts them into speakable time slots before returning them to the voice agent.
Booking Engine
A separate scenario writes the agreed reservation to the calendar with party size and special requests attached, and dispatches automated email/SMS confirmation.
Logging Pipeline
Every call writes a structured row to Google Sheets — outcome, transcript summary, recording link, duration, and cost. This is the exact data layer a custom dashboard is built on.

What This Means for Your Business

How this architecture translates to hospitality and service businesses across the US:

🍳

Answers When Your Team Physically Cannot

Service rush, job site, or treatment room — the constraint is the same. The phone rings hardest when nobody is free to pick it up.

📅

Offers Real Times, Not Callbacks

Checking live availability during the conversation is the difference between booking someone immediately and promising to ring them back.

📝

Captures What People Actually Say

Special requests, unusual party sizes, and specific details a rigid web form would never have asked for.

📊

Leaves a Record of Everything

Every call summarized, recorded, and costed. You can see what your phone line is doing for the first time.

Where This Works, and Where It Doesn't

We are transparent about where voice automation excels and where a hybrid approach is required:

STRONG FIT

This Works Well For

  • Businesses taking bookings by phone during their own busiest hours
  • Reservations, appointments, and table or slot bookings with clear capacity rules
  • Anywhere the person answering the phone is also doing another physical job
  • Businesses that currently keep no record of who called or why
WEAKER FIT

This Is a Weaker Fit For

  • Venues where every booking is negotiated — large private events, custom buyouts, catering quotes
  • Businesses with no defined capacity rules for the system to follow
  • Situations where the call is primarily an exploratory sales pitch rather than scheduling

💡 Hybrid Architecture: For the middle ground, the agent handles routine bookings and queries, and seamlessly transfers large event buyouts to a manager with full context preserved.

BUSINESS ON AUTOPILOT

Want One Built for Your Business?

On the strategy call we'll go through how calls reach you now, what your booking rules actually are, and what an agent would handle. If it's not a fit, we'll tell you.

RESTAURANT PEAK SERVICE EXAMPLE
Friday 7:00 PM Dinner Rush Call Log
19:04
Table for 6 · 8:15 PM Answered in 4s · Special request captured
RESERVED
19:18
Gluten-Free Menu Inquiry Answered from knowledge base
ANSWERED
19:35
Table for 4 · 8:45 PM Calendar checked & booked
RESERVED
20:02
Private Dining Buyout (25 pax) Warm transfer to General Manager
TRANSFERRED
20:40
Table for 2 · 9:15 PM Instant SMS confirmation sent
RESERVED

Prefer to speak directly? Call us: +1 (346) 296-6216

Ready to collaborate with us?

Tell us about your business and we'll design your tailored AI blueprint.