The mistake nobody made

The night of the count

Every year a retailer counts everything it owns and finds less than its records claimed. The difference has a name and a reputation. Shrinkage is assumed to be theft, and to be somebody's fault. This lesson takes a year of it apart, cause by cause. Every cause turns out to have a mechanism behind it rather than a culprit.

Sunday 24 January 2027. Netherby counts every shop after close, and by Tuesday morning Ottoline Tolhurst has the department's year.

Say the basis out loud, because it changes the number by more than half. GBP 162,400 is at retail. It is what the department stopped being able to sell. GBP 68,208 is at cost. It is what left the bank. Both are true. A meeting in which one person quotes each is a meeting about nothing.

Now the part that matters. Netherby can attribute five of the six lines. That is unusual, and it is the product of a year of counting rather than of a clever system.

Read the table again and look for the person who got it wrong. There isn't one.

The largest line is a decision that was correct

Netherby's distribution centre at Wraysholme receipts inbound cartons by scanning the label on the outside. Nobody opens them. When a carton says six packs and holds four, the two missing packs enter the book at the moment of the scan and leave it eleven months later at the count.

That is GBP 47,600 a year, and it is the biggest single line. So Ottoline costed the alternative. Piece-counting every inbound carton at Wraysholme was quoted at GBP 214,000 a year in extra labour. And it would not recover the whole GBP 47,600 either. A realistic detection rate on a manual recount is around seven tenths, so GBP 33,320. Spending GBP 214,000 to recover GBP 33,320 is not a close call. It is wrong by a factor of more than six.

So the largest cause of Netherby's shrinkage is the correct answer to a question somebody asked properly and answered with arithmetic. That is what this lesson is named after. If you go looking for negligence in a shrink number you will find some, and it will not be where the money is.

But the decision is not free in the way it first looks. The reason is about evidence rather than operations. A supplier that short-ships owes for the shortfall. Netherby cannot claim for a carton it never opened, because the claim needs evidence and the scan destroyed the only chance to gather it. The choice was never "count everything or count nothing". It was between counting for detection, which does not pay, and counting for evidence, which might.

Netherby now opens one carton in forty, chosen at random, at a cost of GBP 5,350 a year. It will never find most of the short shipments. Its purpose is to establish a rate per supplier that a claim can be built on, and to make it known that the rate is being established. Detection and deterrence are different products, and they need different amounts of counting.

The bigger number is not on this page

Everything above is the loss that has an account. Alongside it, Ottoline ran the measurement from lesson 2 on the nine fastest-selling styles in the department, counted every four weeks through the sixteen-week autumn phase. Those nine styles did GBP 1,340,000 in the phase, 23.4% of the department's GBP 5,720,000.

Two things in that table, and the second is the one to keep.

First, the size. Shrinkage on those same nine styles over the same sixteen weeks was GBP 11,600. The availability loss was GBP 41,900. That is 3.6 times the shrinkage, on the same styles, in the same shops, over the same weeks. The smaller number has a line in the accounts, a board slide and an owner. The larger one has none of those. And it has none of them for a structural reason rather than a careless one. A sale that did not happen generates no transaction, so no accounting system on earth can see it. You only ever find it by measuring the shelf against the demand curve, and that is work somebody has to choose to do.

Second, the shape. The loss fell from 5.3% to 1.6% across the four windows, down by seven tenths. Nothing changed about theft. Nothing changed about staffing, the product, the ticket price or the weather. The only thing that changed was how long an error was allowed to live before somebody counted it.

What Netherby decided, and how big the bet actually is

The nine-style programme is being extended to the 34 styles that make 58% of department sales, or GBP 10,788,000 of the GBP 18,600,000. Each door counts those 34 styles across their sizes every four weeks. Netherby timed it at 42 minutes a door.

• 62 doors, 13 windows a year, 0.7 hours a count, at GBP 13.60 an hour: GBP 7,673 a year.

Set against that, the measured improvement on the nine styles by the fourth window was GBP 12,300 in a single window.

And here is the honest sentence, which is the one most retail business cases leave out. If the 3.1% held across those 34 styles, the prize would be GBP 334,428 a year, and on the whole department it would be GBP 576,600. Netherby does not write either figure in the paper as a benefit. It measured nine styles. Nine styles is what it can claim. And the fastest-selling nine are the ones most likely to be worst, because volume is what generates keying errors and empty pegs in the first place. Extrapolating the worst nine to the calmest hundred and thirty-one is exactly the arithmetic that makes retail business cases untrustworthy.

The case that was approved says something smaller and much harder to argue with. A year of counting costs GBP 7,673, and on nine styles alone the measured loss fell by GBP 12,300 in a single four-week window. The bet is smaller than one window of the improvement already observed. If that improvement turns out to be the only thing it buys, it has still paid. Take the cheap bet, keep counting, and let the second year's number be a result rather than a hypothesis.

Act as a retail loss and inventory director who has seen a shrink number blamed on the wrong people. Here is my year for one department. Net sales at retail [AMOUNT], book closing stock at retail [AMOUNT], counted closing stock at retail [AMOUNT], blended intake margin [PERCENT]. Whatever cause data I have is here: [PASTE - CAUSE, VALUE, OR SAY I HAVE NONE]. My receiving method is [DESCRIBE - DO WE OPEN CARTONS OR SCAN THEM], my write-off procedure is [DESCRIBE], and my counting programme is [FULL COUNT HOW OFTEN, CYCLE COUNTS ON WHAT]. Do the following. First, compute shrinkage at retail and at cost, state which is which, and express it as a percentage of net sales. Never quote one basis without saying which it is. Second, lay out the causes as a table that SUMS to the shrinkage. Show anything I cannot attribute honestly as unexplained, rather than spreading it across the others. Third, for each cause, name the decision or control that produced it, and say whether that decision looks correct on its own terms. Where I have given you a cost, do the arithmetic on whether fixing it pays. Fourth, tell me for each cause whether counting more often would shorten its life, and estimate the average age of an error under my current programme against a four-week cycle. Fifth, and separately, tell me what I have NOT measured: the sales lost because the book was wrong at store-and-size level. Tell me how to measure it on a small number of high-velocity styles. Be explicit that until I do, it is a hypothesis and not a benefit. Sixth, price a cycle-count programme in store hours at my labour rate, and set it against the smallest measured benefit rather than the largest hoped-for one. Do not tell me my shrinkage is good or bad by comparison with any industry figure. And do not attribute an improvement to anything before checking whether my count method or my sales denominator moved.