What retail knows about its own stock

What the research establishes, the question it leaves unasked, how we answer it, and what a first pilot would produce for both sides.

Issued: 19 October 2026
Author: Alejandro Gil Galue
Founder, Estiba Limited
Reads with: estiba.io/paper
Figures: Measurement of the method, and the cited research

One

The challenge is established, and not by us

Inventory record inaccuracy in retail is not a contested finding or a vendor's talking point. It has been measured repeatedly, at scale, by people with no product to sell.

Nearly 370,000 inventory records were examined across 37 stores of a single retailer. 65% were inaccurate. Variation within a store was explained by SKU-level factors including item cost, selling quantity and method of distribution. DeHoratius, N. and Raman, A. — Inventory Record Inaccuracy: An Empirical Analysis. Management Science 54(4), 2008

Across seven of Europe's largest retailers, around a million SKUs in roughly a hundred stores, about 60% of the SKUs analysed were affected by record inaccuracy. Rekik, Y., Syntetos, A. and Glock, C. — Measuring the Sales Impact of Improving Inventory Records. ECR, 2019

In a UK grocer, across about 24,000 SKUs and 11 stores, average inventory level carried roughly ten times the effect of sales quantity on inaccuracy, and replenishment frequency about 47% more. Perishable lines ran about 6.4% higher. Rekik, Y., Oliva, R., Syntetos, A. and Glock, C. — Inventory record inaccuracy in grocery retailing. Working paper

And the instrument most businesses rely on to find all this is described candidly by the people who sell the systems it sits beside.

“You might not discover it until the end of the year, making it impossible to find the root cause.” Lightspeed, on physical inventory counts

So the position is not in dispute. A majority of records are wrong, the drivers are known, and the instrument used to check them cannot say when or why. None of those three sentences is ours.

Two

The question nobody is asking

Every study above measures how wrong. The question a business actually has to answer is a different one, and the instrument it uses cannot reach it.

Was any of it material?

A margin of error is a property of a measurement you repeat. A count taken once a year produces one observation of each line, and a sample variance carries n − 1 in its denominator, so for any single line there are no degrees of freedom at all. The variance does not come out large. It is undefined.

A full count does produce variance across a catalogue. What it cannot produce is variance within a line, and a line is where a judgement about materiality actually rests.

So “it was within tolerance” and “we have no idea” describe the same state of knowledge at the year end.

The reassurance is produced by the instrument rather than found by it.

This is the challenge we would put to any retailer, and it is the one part of this document that is ours rather than the literature's. You may well be right that it is immaterial. The instrument you are using could not have told you if it were not, which means the decision to absorb it has never actually been tested.

There is a practical version of the same question, and it takes a minute. What share of your fastest-moving lines is not where the record says it is? Most businesses cannot answer it at all. That is not a gap in anybody's arithmetic. It is what an instrument used once a year leaves behind.

Three

How we answer it, and who else has tested the answer

If error concentrates, the counting should too. A set of lines is selected each morning, weighted by what a line is worth, how fast it has been selling and its own history of discrepancies, and counted in the twenty minutes before the doors open by whoever is on shift.

That prioritisation is not our hypothesis. It has been evaluated independently.

Across 150 stores and about 5,000 items, with physical audits run to validate the policies, rule-based policies favouring items with high sales volumes, inventory levels and past errors were the most effective at detecting discrepancies. The best policies uncovered over twice the discrepancies found by random selection, and the policy resembling uniform cycle counting was not a winning one. DeHoratius, N., Holzapfel, A., Kuhn, H., Mersereau, A. and Sternbeck, M. — Evaluating Count Prioritization Procedures for Improving Inventory Accuracy in Retail Stores. Manufacturing & Service Operations Management 25(1), 2023

Two things about that study we would rather raise than have raised with us. Its setting is drugstore and grocery retail, not apparel, where replenishment and stock depth behave differently. And at the 4% counting depth it tested, even the best policy found 8.66% of the inaccuracies present, which says prioritisation is necessary and not sufficient: depth and frequency carry the rest.

What we add to the selection

A session cannot be closed while a difference is unexplained. The person at the shelf is asked what they found, and the reason is chosen by them and never suggested by the system, because a reason that was offered and accepted is not evidence of anything. Most platforms can record an adjustment and most have a field for a reason. A field is optional. A gate is not.

What that produces over a quarter is the thing no stocktake produces: a dated series on the same line, with what was expected, what was found, where it was found, who looked and why they thought it differed.

What we have measured ourselves

83%

of planted decay caught

3.7%

of healthy lines flagged

6,240

line checks a year
against 616 in one full count

18 days

expected wait to see a difference,
against 182.5

The first two come from synthetic books where the answer was known by construction, at a threshold of 1.35 times the comparable rate sustained over two periods. They measure the rule, not a shop. How much work that 3.7% represents depends on how many healthy lines a book holds, which is why the settings belong to the retailer rather than shipping as a default. The second two are arithmetic on twenty lines a day, six days a week, against a 616-line book.

Four

What a first pilot would leave behind

What follows is not a result. It is what each side would hold at the end of two months, and it is worth stating both ways round, because a pilot is a trial of Estiba as much as of anything in your shops.

What you would have

The share of your fastest-moving lines that were not where the record said, counted on dated mornings rather than estimated
A dated series per line, so a repeated correction stops reading as three unrelated ones
Differences separated into found in the wrong place and never found, which are different problems with different remedies
A year-end variance that arrives already explained, line by line, with a date and a name against each one
No new headcount, no integration, no hardware, and systems exactly as they were if you stop

What we would have

A measured result in a working shop rather than in a synthetic book
The threshold and persistence settings that suit a real apparel book, which the literature says cannot be universal
Evidence about whether stock depth belongs in the selection alongside value and rate of sale, which the grocery research suggests and we have not tested

And one condition we would impose on ourselves. The measurement is taken by someone neither party employs. A first reference measured by the supplier is worth nothing to the next buyer who asks, and we would rather have a number that survives the question than one that flatters us.

No saving is claimed anywhere in this document. What is claimed is that at the end of it a business can answer questions it cannot answer today, and that the answers will be in its own numbers rather than ours.

The ask

Twenty minutes, in one of your shops, before it opens

We bring the demonstration; you bring nothing. No stock file, no system access, no preparation. You can tell us whether the lines it picks are the ones you would have worried about.

If it is not for you, say so and we will leave it there.

Start with a conversation →

References

Sources

DeHoratius, N. and Raman, A. Inventory Record Inaccuracy: An Empirical Analysis. Management Science 54(4), April 2008.

DeHoratius, N., Holzapfel, A., Kuhn, H., Mersereau, A. J. and Sternbeck, M. Evaluating Count Prioritization Procedures for Improving Inventory Accuracy in Retail Stores. Manufacturing & Service Operations Management 25(1), 2023, 288–306.

Rekik, Y., Syntetos, A. and Glock, C. Measuring the Sales Impact of Improving Inventory Records. ECR, 2019.

Rekik, Y., Oliva, R., Syntetos, A. and Glock, C. Inventory record inaccuracy in grocery retailing: impact of promotions and product perishability, and targeted effect of audits. Working paper.