Four stores, priced by hand. Now a price change lands everywhere in under two minutes.

Pricing was a four-store scramble. Someone updated the main store on Monday, copied it to the next two on Tuesday and Wednesday, then patched the last one Thursday — and by Friday it had all drifted again because supplier costs moved mid-week. Returns were handled one at a time by a single person, refund fraud was creeping up, and nobody had time to check for it. Their operations chief said the thing blocking growth wasn't demand — it was that pricing couldn't keep up past three stores without breaking.

1. A couple of months of price changes and competitor prices was enough to see a few clear pricing approaches that covered almost the whole catalogue.
2. Each store allows different things, in annoying ways, so we catalogued exactly what was possible where.
3. Six months of refunds, run against known fraud patterns, turned up a small but real slice worth a closer look.

• Their merchandising lead was in the first few tuning rounds — the pricing rules are her judgment written down; we built the engine, she wrote the rules.
• There was a pricing alert channel where she could veto any automatic price change for the first several weeks.
• The returns team was trained on the fraud queue: the system flags, they decide.

• The pricing engine takes competitor prices (we built the part that gathers them), supplier costs, and stock levels, applies the right approach per product, and pushes it to all four stores.
• The store-specific limits are baked in so nothing breaks a rule on any one store.
• Returns are handled automatically: a refund comes in, gets screened for fraud, and is either approved under a threshold (with a record kept) or queued for a person with the warning signs highlighted.
• Competitor prices are checked daily, and only real movement gets flagged — not noise.