The Annual Count Isn't a Stock Correction — It's an Invoice for the Errors

The symptom

In many businesses the annual count follows a fixed ritual. One weekend, extra people, count sheets, discrepancies, recounts, and finally a posting. The difference is noted, usually with a phrase like "within tolerance" or "we've seen worse".

Afterwards the stock figure is regarded as correct — for about a week. Then the same process that produced the difference in the first place starts again.

What nobody calculates: the count difference isn't a snapshot. It's the accumulated inaccuracy of an entire year, and throughout that year the business planned, purchased and committed delivery dates on exactly those wrong numbers.

The mechanism

Stock discrepancies almost never come from theft, though that's the first assumption. They arise at three entirely unspectacular points.

Withdrawals without a booking. A fitter takes two seals off the shelf because he needs them now. The booking happens later, or never. That isn't negligence, it's a rational response to a booking transaction that takes longer than the withdrawal itself.

Bookings without a withdrawal. The production order backflushes the planned quantity per the bill of material, not what was actually consumed. Offcuts, scrap or returns immediately create a deviation nobody notices, because it looks systemically correct.

Wrong storage locations. The material is there, just not where the system thinks. For planning purposes that's identical to "not available" — except that it reappears at the count and offsets the difference elsewhere. Two errors that cancel out in the annual total and both take full effect in daily operations.

The decisive property of all three: none raises an alarm. A wrong stock figure doesn't announce itself. It becomes visible only when somebody stands in front of an empty shelf that the system says is full — and by then it's an acute problem, not a data problem.

The cost

An anonymised example, altered in detail, real in pattern. A wholesaler with around 6,000 stocked articles reported a count difference of about 0.9 per cent of stock value. Internally that was regarded as good.

It got interesting under a different analysis. Instead of looking at the value of the difference, we measured the hit rate per article: for how many articles did the counted quantity match the booked quantity exactly? The answer: 71 per cent. For nearly three articles in ten, the system figure was wrong — just in both directions, so the value largely cancelled out.

The downstream cost sat precisely where the business hadn't looked. Over twelve months, 340 expedited orders were raised, mostly for articles the system showed as available. Average additional cost per expedited order: €62 in freight plus internal effort. On top of that, 84 customer delivery delays traceable to stock discrepancies.

Together, around €31,000 directly attributable — against a count difference worth under €20,000. The difference was the smaller part of the damage.

The fix

The annual count wasn't abolished, but it was relieved of its role as the control instrument.

A rolling cycle count was introduced: every day, one person counts twenty articles, selected by turnover and value. That takes around 40 minutes and covers the A articles several times a year. What matters isn't the correction but that every deviation gets a question about its cause while the cause is still reconstructable. At an annual count it never is.

Second, the withdrawal booking was made faster than the withdrawal itself. Scanner at the rack, one step, no login. As long as booking takes longer than taking, nothing gets booked — that isn't a question of discipline.

Third, returns from production got their own very simple route. Previously they landed, predictably, "for now" on an interim shelf.

After nine months the hit rate stood at 94 per cent and expedited orders had fallen by more than half. The annual count itself took less time, because there was less to recount.

The point is the metric: what counts isn't the value of the difference but the hit rate per article. Value discrepancies cancel out; article errors don't. Metrics like this are exactly what a process analysis surfaces. Fixed scope, fixed duration, fixed price.

The next step

Tomorrow, count twenty random A articles and compare against the system figure. Not the value — the number of exact matches. That single number tells you more about your warehouse than the last annual count.