Deviation

Every retailer keeps two inventories, the one in the system and the one on the sales floor, and in a store more than 60 percent of stock records do not match the counted goods. The deviation is small, a few units up or down, and it is invisible as long as no one counts. Cortex reads system stock, goods receipts, transfers and sell-through in the same time grid and detects from their course where the two inventories drift apart, before the stocktake does.

The deviation comes from goods receipt, the checkout and transfers, from breakage and shrinkage, and from goods that are in the store and not where they are booked, left behind at the service desk or in the wrong slot. For revenue, an item that no one finds is as good as nonexistent. In German retail, errors in recording, posting and valuation alone caused 780 million euros in damage last year, and between two stocktakes the deviation keeps growing.

Standstill

The costly case is the one in which the system records more than is on the shelf. The stock is above the reorder point, so no one reorders, the shelf stays empty, customers move on, and the item disappears from sell-through without an alert being created. It stays that way until someone physically checks.

At every count, between eight and 24 percent of a grocery retailer’s items are affected, and the share grows with the interval between two counts, from around 18 percent with monthly counting to over 27 percent with semi-annual counting. On the other side is the stock that the system records too low, and replenishment that comes too early and ties up capital. In both cases, replenishment planning calculates with a number that does not exist.

Suspicion

Cortex forms the suspicion from the behavior of the checkout. An item that, according to its pattern, sells daily and has not scanned for days while the system lists it is very likely phantom inventory. System stock, last posting, goods receipts, transfers and sell-through lie in the same time grid, and from them a probability emerges per item that the figure in the system is wrong.

Footfall is the context, because an item without sales on a day without customers is no cause for suspicion. Cortex weights the standstill by the footfall in front of the shelf and by the item’s revenue and margin, so that the check begins where it recovers revenue. Shelf sensors confirm the finding where they exist. They are not a prerequisite, any more than RFID or cameras.

Correction

The suspicion goes to the role in the store as a task, with item, shelf location and reason. Check, count, correct the stock, and the order starts again. When incorrect stock records are corrected, revenue rises by four to eight percent, measured in test stores against control stores. The recorded reason feeds into the statistics, so that the cause becomes visible, goods receipt, the checkout or the wrong slot.

Counting is targeted rather than across the board, a few items per day with the highest suspicion instead of a full stocktake that ties up the store for a day. Inventory accuracy and correction time are tracked per location and compared across the network. The stock in the system thus turns from an assumption into a figure that can be brought a little closer to the sales floor every day.