Your Stock Records Are 65% Wrong

At one large retailer, 65% of stock records failed to match the shelf count, and that gap cut profit by more than ten percent [1]. A later check of nearly 370,000 records across 37 stores at a second chain found the same share wrong [2]. These records are where a sales and operations planning (S&OP) run starts, so a perfect forecast still fails if it begins with a false count.

The big picture:

A supply plan balances a future forecast against the stock it expects to hold.

That balance opens with whatever the enterprise resource planning (ERP) software says sits on hand, so any error goes straight into the buying queue. When a record claims too much stock, the system orders nothing and the line waits. When it claims too little, the plan buys what the warehouse already owns.

What strikes me is how this counting gap outlives every new forecast tool on the market. The 2001 study found huge accuracy gaps between stores inside a single chain running the exact same software [1]. The software did not choose which stores kept clean records, placing the true limit in the physical count rather than the planning tool.

A 2008 study then mapped where these errors gather. Of all the spread in record error, 26.4% lived between product groups while just 2.7% sat between stores [2]. Put plainly, how wrong a record gets depends about ten times more on the product line than on the building that holds it.

A 2026 grocery review of roughly 24,000 items across 11 stores explains why. Errors climbed with average stock levels, with how often an item is restocked, and with whether it spoils, but fell with promotional activity [3]. That last detail is the one I did not expect, and it goes against the common story that promo weeks wreck the data.

The cash impact is the part planning teams rarely hear. In that same 2026 study, a shelf count lifted store sales 11% during a field test, and all of that gain landed on items whose records claimed more stock than the shelf held [3]. The count did not sharpen the forecast, but it fixed the false starting line the forecast was matched against.

By the numbers

65% — Records that miss: Sixty-five percent of nearly 370,000 stock records from 37 stores of one retailer were wrong, matching an earlier study of a second chain [2][1].
26.4% vs 2.7% — Where error sits: The share of the spread in counting error that sits between product groups, against the share that sits between stores, so counts belong to groups before sites [2].
11% — Sales lift: The store-wide sales gain from a physical count in a grocery study of about 24,000 items, driven entirely by fixing overclaimed stock [3].
95% to 99% — Planning floor: The record accuracy that material requirements planning (MRP) and ERP systems need to work, where warehouse teams report passing 99% once counting is run properly [5][4].

What I’d watch:

I keep circling back to the order of spend, because each step below costs less than the sensing tool it protects. I am watching which operations teams fix these gaps before they buy a software upgrade rather than after.

Group-first counting: The measured spread puts ten times more of the clear error in the product group than in the physical site, so a count plan built per warehouse targets the smaller share of the problem [2].
Chasing the overclaims: This is where the actual sales gain landed, and it is the error that tells a planner the stock is already covered when the shelf is bare [3].
The active tenth: Counting the 10% of locations that moved since the last check, weighting fast movers and costly items, is the exact method trade press reports using to lift accuracy to 99% [4].
The nearest stage: When physical counts are rare, a two-step supply chain with similar error rates gains more from counting the stage closest to the buyer [6].
A go/no-go number: Below the 95% to 99% accuracy band an ERP run needs, a sharper forecast simply feeds into a broken balance, showing up in the alert queue rather than the accuracy report [5].

The catch:

My read is that a clean record is a needed starting point, not a stand-in for real forecast work.

The academic studies count individual records, not actual dollars. A wrong record on a cheap, slow item and a wrong record on a fast seller are not the same money risk. That is why where the gain gathered, on overclaimed items, matters more than the headline error rate.

That 11% sales lift also came from a single grocery chain running a field test across 11 stores [3]. It belongs to that grocery setting and should not be mapped onto a distribution centre without proof.

A physical count is a snapshot of being right, not a lasting state. Accuracy decays with every single receipt, pick, and return, making the repeated labor cost of counting the true price of the metric. A better record ensures the plan acts on the right stock, but a sensing layer still has to earn its keep by calling what next month will sell.

At a glance

The Big Shift: My read is that the stock record an S&OP plan builds on is the number nobody audits, and it is wrong at a scale that dwarfs the forecast debate, running about ten to one by product group rather than by site.
Why It Matters: A plan that orders against a record claiming stock that is not there ties up working cash in a phantom. Fixing those exact overclaimed records lifted store sales 11% in the study that ran the count.
What I’d Watch:
Group-first counting: Assigning physical counts by product group and item speed rather than evenly per site, targeting where the measured error spread actually sits.
Overclaim bias: Aiming the count at records that claim more stock exists than the shelf holds, which are the items where the sales gain landed.
The accuracy floor: Treating the 95% to 99% record accuracy band as the must-have entry price for a planning upgrade rather than a simple warehouse metric.
The Catch: The figures come from retail and grocery studies that count records rather than money risk, and a fully correct record still does not answer what future demand will do next.

Sources

[1] Raman, DeHoratius & Ton, “Execution: The Missing Link in Retail Operations”, California Management Review 43(3), Spring 2001 — https://cmr.berkeley.edu/2001/05/43-3-execution-the-missing-link-in-retail-operations [2] DeHoratius & Raman, “Inventory Record Inaccuracy: An Empirical Analysis”, Management Science 54(4):627–641, April 2008 — https://ideas.repec.org/a/inm/ormnsc/v54y2008i4p627-641.html [3] Rekik, Oliva, Glock & Syntetos, “Inventory record inaccuracy in grocery retailing”, arXiv:2506.05357v3, 8 June 2026 — https://arxiv.org/abs/2506.05357 [4] Modern Materials Handling, “Rethink Cycle Counting in your Operations”, 9 May 2017 — https://www.mmh.com/article/rethink_cycle_counting_in_your_operations [5] Strategos, “Cycle Counting & Inventory Record Accuracy” (practitioner reference) — https://strategosinc.com/RESOURCES/10-Inventory/CC1-inventory_record_accuracy.htm [6] Kök & Shang, “Evaluation of cycle-count policies for supply chains with inventory inaccuracy and implications on RFID investments”, European Journal of Operational Research 237(1):91–105, 2014 — https://ideas.repec.org/a/eee/ejores/v237y2014i1p91-105.html

Search 133 dispatches