A manufacturer or a distributor does a physical count at year end and finds the system was wrong by a few percent. The response is almost always the same: count more often. Cycle counts are scheduled, a person walks the aisles every week, the variance report grows a tab. Accuracy improves for a while and then settles at a number that nobody is pleased with.
Counting finds inventory errors. It does not make them. The errors are made earlier, at a small number of moments when stock changes hands, and a counting programme that starts at the shelf is always chasing errors that were made weeks ago by people who have forgotten the context.
Stock in a small operation moves through three handoffs on its way to a customer, and nearly every variance can be traced to one of them.
Receiving. Goods arrive and a person enters them. The mistakes here are boring and constant: the purchase order was for cases and the system holds each, the delivery was short and the receipt was entered for the full order, a substitute item arrived under the original's code, or a pallet was received on the dock but not entered until the next morning, after something had already been picked from it.
Put-away. Goods move from the dock to a place. The record says the shelf, the pallet is in the overflow area, and the next picker looks in the wrong place, concludes it is out of stock and short-ships or raises an emergency purchase. The stock was never lost. Its location record was.
Picking. Goods move from a place to a shipment. The picker takes a nearby item that looks the same, or takes the quantity on the paper rather than the quantity on the shelf, or breaks a case and does not record the remainder. Returns arrive back through the same door and are put away without a transaction.
Each of these is an ordinary moment in an ordinary day, and each produces a small discrepancy that grows silently until the next count finds it. By then the context that would explain it is gone: who received that pallet, what the label said, which picker was on the shift.
A count is a measurement after the fact. It tells you that a gap exists and how big it is, and it can tell you which item. It cannot tell you why, because the handoff that created the gap happened days or weeks before and left no trace. The team adjusts the record to match the shelf, writes "count variance" as the reason, and the cause is lost.
That is why accuracy plateaus. The adjustments make the numbers right and leave the three handoffs exactly as they were, so the same errors recur at the same rate and the next count finds the same size of gap.
We described the related problem of systems keeping separate counts in three counts, none of them right: when a storefront, an accounting system and a warehouse each keep their own number. This is the physical side of the same issue. Even with one system, the number is wrong if the handoffs are loose.
The alternative is to move the measurement to where the error is made. Instead of a calendar of counts, use events that trigger a count, one for each handoff.
At receiving: count a sample of every receipt against the document before it is entered. Not every line, a sample, chosen by a rule such as the first three lines and any line with a new supplier or a unit of measure that changed. The check takes a few minutes and catches the short delivery and the cases-versus-each error while the truck is still there.
At put-away: confirm the location by scan or by a second person for any item stored in more than one place. The cost is a few seconds per pallet. The benefit is that "not in stock" stops meaning "not where the system says".
At picking: count what remains after any pick that breaks a pack or empties a location. If a picker takes the last four from a location, the location is confirmed empty or the remainder is counted, immediately. This is the cheapest count in the building, because the person is standing there.
Each check creates a record with the reason attached: the receipt, the location, the pick. When a variance appears later, it can be traced to the last confirmed state at a handoff, and the cause is a short search instead of a guess.
Accuracy needs a number you can track by handoff, not one number for the warehouse. Three ratios are enough: the share of receipts that matched the document in the sample, the share of put-aways confirmed at the first location check, and the share of picks that left a correct remainder. A fall in any of the three points to the handoff, and usually to the day, the shift or the supplier.
The point of splitting the measure is that each ratio has an owner. Receiving belongs to the dock lead, put-away to the warehouse lead, and picking to whoever manages the pick floor, so a fall in a ratio lands on a person who can change the process that day.
Annual and cycle counts remain, with a different role. They become the audit of the handoff controls: if the controls work, the count finds little. A count that finds a lot is the signal to look at which handoff slipped, not a reason to schedule more counts.
The saving is not mainly in the adjustments. It is in the decisions made on bad data between counts: the purchase order raised for stock that was on the shelf, the sale lost because the system said none was available, the customer promised a ship date on stock that had already been picked for another order. Those costs are real and invisible, because nothing in the books is labelled "decision made on a wrong number".
There is a second saving, which is trust. A team that knows the numbers are checked at each handoff stops keeping a private spreadsheet of what is really on the shelf. That spreadsheet is the usual sign that the system of record has stopped being one.
Manufacturers have one more: the moment raw items become a finished item. A bundle or assembly consumes several lines and produces one, and if nobody decides when that conversion is recorded, the components are counted twice, once as parts and once inside the finished item, or not at all. We walk through it in one assembly, three stock counts. The control is the same as at the other three handoffs: confirm at the moment it happens, record the reason, and give the conversion a single owner.
Start with the handoff that causes the most trouble, which the team can usually name without looking at any data. Run the matching check for two weeks, record every exception with its reason, and compare the exceptions with the variance the next count finds. If the two overlap, you have located the leak. If they do not, you have learned the error is made somewhere else, and that is also worth knowing.
We design the three handoff controls with the people who work the dock and the floor, and connect them to the inventory and accounting records so the exceptions reach one place. Discovery is no-risk: we trace last quarter's variances to their handoff and show you where the leak is, and you pay only if you go ahead. The build carries a guaranteed estimate, so if we estimate low, we absorb the difference.
Accuracy is not something a count achieves. It is something the handoffs either keep or lose, and the count only reports which.
In a no-risk discovery we follow an order from sale to shipment to invoice across your systems and show where it breaks and what one connected system would change. You pay only if you proceed. Or see how we approach it.
More on the same problem:
"Something isn't working." "We're outgrowing our tools." "We're afraid to make the wrong move." No-Risk Discovery is a short, practical conversation that gets you clarity before you commit to anything big. We'll tell you if we're a fit. If we're not, we'll tell you that too.
Say who you are and ask. A person replies within one business day.
