60% of all retail stock records are wrong, shows new research

A new study of 1.3 million stock audits across six major retailers finds that most inventory record errors are predictable, and the most damaging ones can be caught early.

The report from ECR Retail Loss outlines a new stock-taking model that finds real errors nine times out of ten
The machine learning model, built only from data retailers already hold, catches around 19% more errors than the best conventional method
It flags over 80% of “phantom” stock, the empty shelves the system wrongly believes are full, before the lost sales mount up.
Around 60% of all stock records in retail are wrong, creating a disparity between what the system says is in the storeroom and what’s on the shelf.

New research from ECR Retail Loss provides a solution for retailers looking to bridge that reality gap and help ensure the right product is available at the right time.

Retail’s traditional fix is to count more and count more often. However, counting is slow, costly and increasingly hard as ranges grow and staffing gets tighter.

The report, by Professors Yacine Rekik, Aris Syntetos and Christoph Glock, examined more than 1.3 million stock audits collected between May 2018 and April 2022.

And the authors conclude that the most effective approach is not to count everything, but to count smarter.

Their new report, Smart Inventory Record Inaccuracy (IRI) Prediction and Management, shows that most stock errors are predictable, and outlines a model to tackle them.

Using a transparent machine learning model, built on data retailers already hold, the model can forecast where records are drifting and rank the day’s checks accordingly, or even correct them automatically where it is certain the stock is not there.

Stable, high-volume lines can be safely set aside, while volatile categories, such as fresh, short-life and heavily promoted products, get more attention.

The biggest prize for retail is phantom inventory, where the system believes a product is in stock but the shelf is empty. And a reorder only fires when staff check by hand.

The new model flags more than 80% of these cases early, before the lost sales mount up.

Counting will not disappear in this model. Statutory stocktakes remain the anchor, and the least predictable lines need checking more often.

The opportunity is to direct the same effort where it delivers the highest returns.

Notes to editors

Smart Inventory Record Inaccuracy (IRI) Prediction and Management is a report from ECR Retail Loss authored by:

Professor Yacine Rekik (emlyon business school, France)
Professor Aris A. Syntetos (Cardiff Business School, Cardiff University)
And Professor Christoph H. Glock (TU Darmstadt, Germany).
The study draws on more than 1.3 million stock-audit observations collected between May 2018 and April 2022 across six grocery retailers of varying size, format and country.

All performance figures come from a held-out test set of future data the models had not seen during development.

Download the full report here: https://ecrloss.com/research-paper/predicting-inventory-reco...

If you want to speak to the researchers, please contact Peter Remon at BlueSky Education - peter@bluesky-pr.com