Inventory accuracy: the gap between a number and the shelf

Measure where the work happens

Comparing a product's total against a physical count hides offsetting errors, because stock missing from one position and surplus in another can cancel out perfectly. Counting by location, and treating any position whose quantity is wrong as a failure regardless of the size of the difference, produces a much harsher figure and a far more useful one. It is the only measure that correlates with whether picking will run smoothly tomorrow.

Where the drift comes from

The usual causes are mundane and repetitive: a move confirmed without scanning the destination, a damaged case removed and never recorded, a part-picked case left in a different place, a mixed pallet accepted at goods-in without being broken down, an unlabelled return placed on a shelf because the receiving lane was full. None of them feels significant at the moment it happens, and each creates a discrepancy that only a count will find.

Detection has to be built in, not bolted on

A system that permits negative stock, allows quantity edits without a reason, or lets an operator skip a location confirmation will hide errors indefinitely. A system that blocks impossible states, raises an exception when a picker reports an empty face, and queues those exceptions for someone to work through, surfaces problems within hours. The design question is how quickly an error becomes visible, because an error found the same day usually still has an explanation attached.

Adjustments need a reason and a name

Correcting a quantity without recording why converts a symptom into a permanent mystery and removes any chance of fixing the cause. Reason codes, ideally short and honest ones such as damage, mis-pick, unrecorded receipt or found stock, turn adjustments into a diagnostic dataset. Reviewing that dataset by area, by shift and by product is what separates sites that fix causes from sites that count endlessly.

The consequences run past the warehouse

Unreliable records generate emergency purchases, unnecessary safety cover and promises to customers that cannot be met, and the planning teams that consume the data have no way of knowing which figures to distrust. Where goods are held under a customs or duty-suspension arrangement, an unexplained difference has consequences beyond the commercial ones. The replenishment and stock-holding calculations that sit on top of the data belong to planning; the truthfulness of the data belongs to the floor.

Frequently asked questions

Why does a site with high reported accuracy still short orders?

Because the measure is probably aggregate. Errors in opposite directions cancel at product level while both positions remain wrong, so pickers meet empty faces even though the headline figure looks reassuring.

What is the fastest way to improve accuracy in a struggling site?

Close the ways records can be changed without evidence: enforce destination scanning, block quantity edits without a reason code, and make an empty pick face raise a count task immediately. Those three controls remove a large share of everyday drift before any counting programme is redesigned.