How a Mississauga restaurant cut 12 hours/week off scheduling with AI

The starting point

A family-owned restaurant in Mississauga with 15 staff, two managers, and around 200 covers a night. Scheduling lived in a shared Google Sheet. Reservations came in through a mix of phone, Instagram DMs, and a basic OpenTable account. End-of-day reports (what sold, what's running low, what tomorrow looks like) were assembled manually by whichever manager was closing.

When we mapped their week, three patterns showed up over and over:

• About 5 hours per week building and revising weekly shift schedules, mostly via text messages chasing staff availability.
• Roughly 6 hours per week handling reservation phone calls, many of them simple yes/no availability checks that could happen without a human.
• Around 4 hours per week compiling end-of-day notes for the next morning's huddle, almost always after midnight when judgment is worst.

Total: ~15 hours per week of high-friction admin, spread across two managers. That's nearly two full shifts of management time burned on coordination, not cooking, not service, not training.

The three workflows we built

1. AI-drafted shift schedules with one-click confirmations

Staff submit availability through a simple form (Google Forms, no new app to learn). Each Sunday morning, a Make.com scenario pulls availability, last week's hours worked, and a small set of operational rules (minimum cooks per shift, max consecutive doubles, etc.), then asks Claude to generate a balanced draft schedule.

The manager reviews the draft in 10 minutes instead of building it from scratch in 3 hours. Once approved, individual shift confirmations go out to each staff member by SMS via Twilio. Replies ("can't do Friday") flow back into a queue the manager handles in one batch.

2. Reservation triage on Instagram, WhatsApp, and email

An incoming message, whether DM, WhatsApp, or email, gets read by a Claude-powered classifier. If it's a simple availability check ("do you have a table for 4 at 7 on Saturday?"), the system checks OpenTable, replies with a confirmed booking link or two alternatives, and logs the conversation. If it's anything more nuanced (a large party, an allergy question, a complaint), it gets flagged in a Slack channel for a human to handle within minutes.

Result: about 70% of reservation messages now resolve without anyone touching them, and the remaining 30% land in front of a manager already triaged.

3. Auto-generated end-of-day report

At close, the workflow pulls sales data from the POS export, current inventory levels from the kitchen's checklist, and any reservations on the books for tomorrow. Claude turns it all into a one-page plain-English summary that lands in the owner's inbox by 1am: top sellers, dead stock, what to prep, and any red flags.

What used to be 30–45 minutes of post-shift typing is now zero minutes. The owner reads the summary with morning coffee instead of fighting through it at midnight.

What the numbers looked like after 6 weeks

Why this works for restaurants specifically

Restaurant operations are deeply repetitive: the same scheduling problem every week, the same reservation questions every shift, the same end-of-day summary every night. That repetition is exactly what AI workflows handle well. The hard part isn't the AI; it's connecting the systems you already use (POS, scheduling sheet, messaging apps) so that data can actually flow between them.

Most GTA restaurants we talk to don't need new software. They need the existing software to talk to each other, with AI handling the judgment calls in between.