Skip to content

Case study · Hospitality & reservations

No-shows down 83.1%. Fees down to $0.

How a 110-seat bistro cut weekend no-shows by 83.1% and reclaimed $1,420 a month in fees. We replaced manual WhatsApp bookings and a commission-heavy aggregator with a custom reservation engine that confirms in seconds, locks tables atomically and refills cancellations on its own.

Industry
Hospitality & automated reservations
Deployment
3 weeks
Tech stack
React, Tailwind CSS, FastAPI, PostgreSQL, Redis, WhatsApp Cloud API
Core result
No-shows 18.4% → 3.1% · 3.4s confirmations · $0 in cover fees

Verified 60-day results. The baseline is the 30 days before launch; the post-launch figures are days 1 to 60 on the Reozix engine.

Client
A 110-seat urban casual dining bistro with 28 indoor tables and 2 patio zones
Challenge
Weekend no-shows, slow manual replies and per-cover aggregator fees eating prime-time margin
What we built
A React booking widget, FastAPI engine, Redis slot locking and a two-way WhatsApp loop
Outcome
No-shows down 83.1%, 3.4s confirmations and $17,040 a year in fees saved
Deployment scorecard. Weekend no-show rate fell 83.1%, from 18.4% to 3.1%. Booking confirmation takes 3.4 seconds, down from 18 minutes. Annual fee savings are $17,040. Mobile LCP improved from 4.6 seconds to 0.8 seconds. A line chart shows the no-show rate falling from 18.4% before launch to 3.1% by week 8 while Friday and Saturday seat utilisation rises from 74.0% to 88.5%, with the WhatsApp trigger enabled in week 2.

The operational bottleneck

Why manual reservations and aggregators were bleeding margins.

Before partnering with Reozix, this 110-seat urban casual dining bistro was losing prime weekend revenue across three broken reservation channels. Each one leaked tables, money or both.

How it works

From booking tap to refilled table in 4 steps.

One automated loop, with no app download, no password and no front-of-house typing.

  1. 01

    Pick a slot

    The diner chooses a party size and time. The table is locked for 5 minutes in 42 ms.

  2. 02

    Instant pass

    Within 3.4 seconds a WhatsApp pass arrives with a 1-tap calendar file and map pin.

  3. 03

    T-3 hour check

    A Confirm Table or Cancel / Modify message goes out three hours before seating.

  4. 04

    Auto backfill

    A cancelled table is offered to the top three waitlist parties with a 1-tap claim link.

1 · The problem

Three broken reservation channels.

Live service exposed where every booking route was failing the restaurant.

01

The manual WhatsApp & phone trap

Roughly 65% of local diners booked by WhatsApp DM, Instagram message or phone. Between 6:30 PM and 9:30 PM on Fridays and Saturdays, staff were seating guests and could not watch their messages. Median response time stretched to 18 minutes, by which time high-intent groups of 4 to 8 had booked elsewhere. Copying messages into a spreadsheet also caused 3 to 4 double-booked tables every month.

02

The third-party commission tax

The other 35% of bookings came through an aggregator that charged a flat monthly software fee plus $1.50 to $2.50 per seated cover. It charged even when diners clicked the booking link from the restaurant's own Instagram bio or Google Business Profile.

03

The ghost table problem

With no pre-arrival verification and no instant waitlist backfilling, 18.4% of weekend reservations failed to show. Prime four-top tables sat empty for 45 minutes while walk-in guests were turned away at the host stand.

2 · The Reozix architecture

A sub-second, zero-commission booking pipeline.

Instead of forcing diners to download a bloated app or register with a password, we built a lightweight React and Tailwind CSS booking widget on a high-concurrency FastAPI backend, with Redis slot locking and a two-way WhatsApp Business Cloud API loop.

A · Slot locking

Concurrency-safe table allocation.

Standard WordPress form plugins fail when two diners try to book the last 8:00 PM four-top at the same second. Reozix added an atomic Redis lock, so the moment a diner picks a party size and time, that table combination is held for 5 minutes. Turn-time rules set seating duration by party size, so the kitchen never gets a spike of tickets at once.

  • Atomic Redis lock with a 5-minute TTL
  • POST /api/v1/reservations/lock returns in 42 ms at p95
  • 2 guests = 75 mins, 6+ guests = 120 mins
  • 10 minutes padded in for table bussing
B · WhatsApp loop

Two-way WhatsApp confirmation and waitlist backfill.

Generic email confirmations sit unread in promotional tabs. Reozix connected directly to the WhatsApp Business Cloud API with three automated triggers across the booking lifecycle.

  • T-0 sec: an instant booking pass with a 1-tap calendar file and Google Maps pin
  • T-3 hours: interactive Confirm Table or Cancel / Modify quick-reply buttons
  • No reply within 90 minutes of a peak slot, or a Cancel tap: the table flips to AVAILABLE and a claim link goes to the top three waitlist parties

Financial ROI

Where $5,500 a month in net value comes from.

Aggregator fees eliminated

$1,420 / mo

Every booking is now 100% direct. No per-cover charges and no monthly platform fee.

$17,040 a year saved

Prime-time tables recovered

$3,180 / mo

Fewer no-shows, and cancelled tables refilled from the waitlist before service.

86% of cancellations refilled

Front-of-house labour reclaimed

$900 / mo

Admin coordination fell from 14 hours a week to 1.5 hours.

12.5 staff hours a week

Combined net impact

$5,500 / mo

Full payback of the custom deployment in 46 days.

6.5 weeks to payback

60-day verified results

1,840 direct reservations, 100% uptime and zero double-bookings.

By day 60 the restaurant had processed 1,840 direct reservations (about 6,100 seated covers) through the Reozix engine.

Two charts. Left: mobile checkout completion rose from 58.4% to 87.2%, the 3-hour reminder open rate rose from 19.2% to 91.4%, and peak table utilisation rose from 74.0% to 88.5%. Right: verified monthly value of $5,500, made up of $3,180 recovered no-show revenue, $1,420 aggregator fees saved and $900 front-of-house labour saved.

3 · Technical data made simple

Under the hood of a sub-second booking engine.

The diner widget, the FastAPI engine, the Redis lock, the PostgreSQL table state and the WhatsApp loop each do one job, and together they keep every booking safe and fast.

42 ms

Reservation lock latency, p95 (vs. 1,280 ms)

An atomic Redis lock replaces a slow database check, so two diners can never take the same table.

91.4%

WhatsApp reminder response (vs. 19.2%)

Interactive quick-reply buttons get answered where email open rates sat at 19.2%.

0

Double-bookings in 60 days

Down from 3 to 4 double-booked tables every month under the manual spreadsheet.

Architecture diagram. 1: the React and Tailwind diner widget, LCP 0.8 seconds on 4G, sends a POST lock request to 2: the FastAPI engine with async Python workers and an adjacency solver, p95 latency 42 ms. The engine uses 3A: a Redis slot lock with an atomic 5-minute TTL and zero double-bookings, and 3B: a PostgreSQL database for ACID table state and combinable tables. Both feed 4: the WhatsApp API with an instant pass, a T-3 hour 1-tap confirm and an auto-waitlist blast, 91.4% open rate. Production telemetry: 1,840 direct bookings, over 6,100 covers seated, 86% of cancelled slots refilled and 100% uptime.

Verified 60-day performance: baseline vs. post-launch

Core operational / technical metric30-day pre-launch baseline60-day post-launch (Reozix)Verified net impact
Weekend prime-time no-show rate18.4% (nearly 1 in 5 tables)3.1%-83.1% reduction
Median booking confirmation speed18.0 mins (manual DM reply)3.4 seconds (automated webhook)99.6% faster response
Third-party aggregator cover fees$1,420 / month ($1.50-$2.50 per cover)$0 / month (100% direct)$17,040 / year saved
Friday & Saturday seat utilisation74.0% capacity88.5% capacity+14.5% peak capacity
Cancelled tables refilled via waitlist14.0% (manual phone calls)86.0% (auto WhatsApp blast)+72.0% recovery lift
Front-of-house admin coordination14.0 hours / week1.5 hours / week12.5 staff hrs/wk saved
Mobile booking widget LCP (4G)4.6 seconds (heavy iframe)0.8 seconds (native React)+28.8% checkout conversion
Reservation lock API latency (p95)1,280 ms42 ms (Redis atomic lock)0 double-bookings in 60 days

4 · The messy middle

What failed in week two, and how we fixed it.

Not every deployment works on day one. Live restaurant traffic exposed two edge cases during weeks 1 and 2 that needed immediate engineering pivots.

Pivot 1: email reminders were ignored (19.2% open rate). To keep messaging costs near zero at rollout, email was the primary T-3 hour reminder with an SMS fallback. Open rates sat at 19.2% and the no-show rate only fell from 18.4% to 15.2%. In week 2 we moved the primary trigger to interactive WhatsApp Business API templates with quick-reply buttons. Response rates jumped from 19.2% to 91.4% overnight, and no-shows fell to 3.1%.

Pivot 2: rigid table combinatorics blocked 6-person parties. During Friday peak service the engine rejected two 6-person reservations because both dedicated 6-tops were booked, even though two adjacent 4-tops (Tables 12 and 14) were free. We updated the PostgreSQL schema to support adjacency groups (combinable_table_ids). The solver now merges two adjacent 4-tops into an 8-seat block at peak demand and splits them back into 4-tops once the turn completes.

5 · Verified outcomes

What the numbers mean for the restaurant.

Three results that moved the business, measured over the first 60 days.

83.1% drop in weekend no-shows, $3,180 a month recovered

Prime-time no-shows fell from 18.4% to 3.1%. 9.2% of guests used the 1-tap WhatsApp button to cancel ahead of time, and 86.0% of those cancelled tables were refilled from the waitlist before dinner service, recovering an estimated $3,180 a month.

Sub-second Core Web Vitals, +28.8% mobile conversion

A lightweight React bundle instead of a third-party iframe cut mobile LCP from 4.6s to 0.8s and INP from 340ms to 48ms. Mobile checkout completion rose from 58.4% to 87.2%.

6.5-week payback on $5,500 a month

$1,420 in eliminated fees, $3,180 in recovered tables and $900 in reclaimed labour (12.5 hours a week) returned the full cost of the custom deployment in 46 days.

FAQ

Questions restaurants ask us.

How does automated WhatsApp confirmation reduce restaurant no-shows?

Unlike email reminders with 19.2% open rates before dinner service, interactive WhatsApp Business API messages achieve 91.4% visibility. Sending a 1-tap Confirm or Cancel button 3 hours before seating captures early cancellations and automatically backfills the table from the waitlist.

How do custom booking systems prevent double bookings during peak hours?

Reozix uses atomic slot-locking via Redis with a 5-minute TTL (42ms p95 latency) alongside ACID-compliant PostgreSQL transactions and combinable table adjacency logic.

Do diners need an app or an account to book?

No. The booking widget is a lightweight React and Tailwind CSS page with direct checkout, so guests never download an app or register with a password. They book, then receive a WhatsApp booking pass.

How long did the build take?

This restaurant booking engine was deployed in 3 weeks, and the verified results cover the first 60 days after launch.

Technology

  • React
  • Tailwind CSS
  • FastAPI
  • PostgreSQL
  • Redis
  • WhatsApp Business Cloud API

Next step

Want a booking system like this for your restaurant?

If no-shows, slow replies or platform fees are costing you tables, we can build a direct booking engine that fixes it.