Payout Auditor: an AI coworker that recovers restaurants' delivery-app money
The brief: visit restaurants that sell on Swiggy and Zomato, pick one problem area, and turn what you find into the AI coworker you'd build. I picked Operations: money lost between an order and the bank credit.
What I found
Three single-outlet restaurants together take about 240 orders a day, 60% of them through Swiggy and Zomato — ₹14.67 lakh a month of order value that reaches the bank only after commissions, discounts, taxes, refunds and adjustments. In all three, nobody could connect an order to the rupees finally credited for it. Payouts were checked only when a credit already looked wrong, and disputes were chased by hand, one order at a time.
Why Operations
- It's money back, not money forward: recovered deductions go straight to profit with no change in customer behaviour.
- It showed up everywhere: all three restaurants had it, across different cuisines, volumes and owners.
- It can be proven: each settlement can be recomputed order by order, so every recovered rupee is traceable.
The coworker: Payout Auditor
- Collects daily orders, the platform settlement report and the bank credit.
- Recomputes what every order should have paid from the agreed commission, discounts, taxes and charges.
- Flags each gap: unexplained deductions, wrongly charged refunds and cancellations, and missing orders.
- Prepares the dispute with order ID, expected versus paid amount and a ready-to-send message, then tracks it to resolution.
- Sends a one-page weekly report: gross to net by channel, money recovered, disputes still open.
How I did it
Short, structured conversations of 20–30 minutes with each owner or manager, logged in a working sheet with formula-driven numbers. I used Claude to structure findings, rank problems, check the arithmetic and draft the report. Restaurants are anonymised and no bank, tax, login, staff or customer details were kept.