Inventory and reorder agent

Background

The client is the head office of a bakery franchise with 60 stores. Each store receives 350 ingredient and packaging items, such as flour, butter, cream, and boxes, from the head office distribution center and two designated suppliers. Ordering was the store owner’s job: after closing, walk the storeroom, count stock, decide quantities by feel (“same as last week”), and send orders to each supplier by fax or messenger. It took 40 minutes a day and was skipped on busy days.

The results cut both ways. With weekend promotions and weather changes unaccounted for, key ingredients ran out more than three times a week per store, while ingredients ordered generously passed their use-by dates and 4% of monthly sales went to waste. Head office knew store stock only from a spreadsheet each owner sent once a month, so distribution center planning ran on numbers a month old.

What we built

The goal was to remove counting, calculating, and copying and leave the owner with only checking quantities and approving. No order goes out without the owner’s approval. We built five things.

1. Stock estimation

Current stock per item is estimated from POS sales, ingredient usage per recipe, and delivery records. Once a week the owner does a count in the app, which corrects the estimate, and large differences prompt for a reason (unrecorded waste, recipe change, and so on). The nightly storeroom walk is gone.

2. Reorder points

For each item, the agent combines sales trend, weekday pattern, head office promotion calendar, weather forecast, supplier lead time, and minimum order quantity to work out when and how much. Use-by dates are tracked per delivery lot; if stock is about to expire, the order is reduced and the owner is told which items to use first.

3. Draft orders and owner approval

When stock falls below the reorder point, draft orders per supplier appear in the owner’s app. The owner sees quantities with reasoning (“weekend promotion, cream demand expected at 1.6×”), adjusts or approves as is. No order goes out without approval. If the owner has not responded within half a day, a reminder is sent, and then head office is notified.

4. Ordering portal

The distribution center and both suppliers receive orders and register deliveries in the same portal. The owner confirms quantities on receipt in the app, and any difference between order and delivery is shown to both sides the same day. Orders that used to travel by fax and messenger now live in one place.

5. Head office dashboard

Stock, orders, waste, and stock-outs across all 60 stores in real time. Items and stores with frequent stock-outs and items with high waste are ranked, and store order forecasts feed directly into distribution center planning.

How it was rolled out

Weeks	Work
1–3	Site visits to 5 stores. Recipe usage, lead times, and minimum order quantities for 350 items. Analysis of a year of POS and order data
4–8	Built stock estimation and reorder points. Four weeks in shadow mode at 5 stores alongside owners’ own orders
9–12	Owner app approval flow, ordering portal, supplier integration. Pilot extended to 15 stores
13–16	Head office dashboard. Rollout to all 60 stores, owner training, handover to head office operations

Over the four shadow weeks, following the agent’s proposals as given would have cut stock-outs by 68% and waste by 21%. Actual figures in the first two months after full rollout were 87% and 23%. The share of drafts owners adjusted fell from 31% in the first month to 9% in the third.

How a store owner’s day changed

Before, the owner’s closing routine ended with 40 minutes of counting stock and sending orders to three suppliers. Now the owner checks and approves the draft orders in the app in under five minutes in the afternoon. Items whose quantities changed because of a weekend promotion or rain forecast carry the reason, so the decision to adjust is immediate. The weekly count covers only items that differ.

Instead of waiting for a monthly spreadsheet, the head office merchandising team sees on the dashboard which ingredients run out at which stores, and matches distribution center receipts to store order forecasts.

The table below compares each task before and after.