The Problem
A 5-branch restaurant chain in Karachi was managed by an owner who started his day with 45 minutes of phone calls. calling each branch manager to ask about yesterday's sales, today's stock levels, and what staffing issues were brewing. By mid-morning his three group chats had 400+ messages he was scrolling through. Food waste was running at 22% of cost of goods because each branch ordered conservatively to avoid stockouts and ended up dumping product.
The owner didn't want a fancy POS replacement. his existing POS worked fine per branch. He wanted the daily picture, automatically.
What I Built
A lightweight ops automation that sits on top of the existing per-branch POS systems:
1. Nightly data extraction. A scheduled job pulls each branch's daily sales, inventory snapshot, and waste log from the POS exports, normalises them, and writes to a central Postgres database.
2. 7am owner brief. A WhatsApp message arrives at 7am with: yesterday's sales by branch, top 5 selling items, locations with low stock, locations with unusual waste, and any anomalies the AI flagged worth attention. Generated by Claude from the previous night's data.
3. Predictive food waste model. A simple time-series model (per-branch, per-SKU) predicts the next 7 days of demand. The system surfaces "your usual order of fish for Branch 3 looks 18% too high based on the last 4 weeks of weather + bookings."
4. Cross-branch alerts. If one branch is out of stock on a popular item, the system flags whether an adjacent branch has surplus that could be redistributed.
Outcome
In 6 months, food cost dropped from 50% to 36% of revenue. adding roughly PKR 4.2M/month to net margin across the 5 branches. The owner's mornings became a 4-minute read of the WhatsApp brief. He repurposed the saved time into negotiating better supplier contracts and scoping a 6th location.