Predictive stock counting: A smarter way to track inventory

Predictive stock counting: the inventory answer you’ve been waiting for

Retailers have spent decades trying to answer the same question.

How do you keep inventory records accurate without spending more time counting stock?

It’s a problem that affects every part of the business.

When inventory records drift away from reality, shelves sit empty even though the system says products are available.

Replenishment orders fail to trigger, colleagues waste time searching for stock that doesn’t exist, and customers leave disappointed.

One of the most damaging examples is phantom inventory, where the system believes stock is available even though the shelf is empty.

Because the inventory record appears correct, no replacement order is generated until someone physically discovers the problem.

The traditional solution has been straightforward.

• Count more.
• Count more often.
• Count everything.

But what if retailers have been solving the wrong problem?

New research from ECR Retail Loss suggests there is a far more effective approach: predictive stock counting.

Academics from Cardiff Business School, emlyon Business School and the Technical University of Darmstadt developed our new approach.

Instead of treating every product as equally likely to contain an inventory error, retailers can use machine learning to predict which records are most likely to be inaccurate before they begin counting.

The result is a smarter, more targeted way to improve inventory accuracy while making better use of scarce labour.

Why traditional stock counting is reaching its limits

Inventory accuracy has never been more important.

Retailers depend on reliable inventory records to replenish shelves, fulfil online orders, forecast demand and understand where losses occur.

Yet maintaining accurate records has become increasingly difficult.

• Product ranges continue to expand.
• Stores process thousands of transactions every day.
• Promotional activity creates rapid changes in demand.
• Labour shortages leave fewer hours available for manual stock checks.

As a result, inventory records naturally drift over time. Small discrepancies accumulate until the system no longer reflects what is actually on the shelf.

Traditional cycle counting helps identify these issues, but it relies on one important assumption.

That every item deserves roughly the same level of attention.

The new research challenges that assumption.

“As a data scientist, it has always been my belief that the potential for machine learning in the field of inventory management has been underestimated,” says Stefan Ahrens, of German retail chain dm-drogerie.

“This research proves that it can bring substantial value and actionable insights to support decisions around stock counting and replenishment strategies.”

What is predictive stock counting?

Predictive stock counting turns the traditional stock-counting model on its head.

Instead of asking: “What should we count today?”

Retailers ask: “Which inventory records are most likely to be wrong?”

Using historical inventory data already held by retailers, predictive models identify products whose inventory records are drifting away from reality.

Rather than checking thousands of items across an entire store, colleagues can focus their efforts on the products where an inventory error is genuinely likely.

Stable, predictable products require less attention.

Volatile categories receive more.

This targeted approach allows retailers to discover more genuine inventory errors while carrying out fewer unnecessary checks.

Perhaps most importantly, predictive stock counting does not require retailers to install expensive new hardware or collect new forms of data.

The research demonstrates that existing operational data can provide enough information to prioritise inventory checks far more effectively than conventional methods.

Using your own data to recognise inventory problems

At first glance, predicting inventory errors sounds complicated.

In practice, the principle is surprisingly straightforward.

Every product leaves behind patterns.

• Some stock records remain consistently accurate.
• Others become unreliable after promotions.
• Some categories experience frequent theft.
• Others suffer from waste or handling errors.

Products with short shelf lives behave differently from long-life grocery lines.

Machine learning is exceptionally good at recognising these patterns across thousands of products simultaneously.

Instead of relying on fixed rules, predictive models continually assess the likelihood that an inventory record no longer reflects reality.

That allows retailers to rank inventory records according to risk, directing colleagues towards the products most likely to contain errors.

Rather than replacing humans, predictive stock counting helps stores direct resources where they will have the greatest impact.

The evidence for predictive stock counting

The research behind predictive stock counting is substantial.

Rather than relying on simulations or small pilot studies, Professors Rekik, Syntetos and Glock analysed more than 1.3 million stock audits collected across six grocery retailers over four years.

The developed machine learning model correctly identified genuine inventory errors around nine times out of ten.

It outperforms the best conventional approach by approximately 19%.

Perhaps even more significant, it successfully identified more than 80% of phantom inventory cases, allowing retailers to intervene before empty shelves translated into lost sales.

These findings suggest that retailers no longer need to choose between inventory accuracy and operational efficiency.

With predictive stock counting, they can improve both simultaneously.

Traditional stock counting vs predictive stock counting

Traditional Stock Counting	Predictive Stock Counting
Counts products according to a fixed schedule	Prioritises products most likely to contain inventory errors
Treats every product as equally important	Focuses effort where it will have the greatest impact
Requires significant labour	Makes better use of existing labour
Finds errors after they have occurred	Predicts where errors are likely to occur
Can miss phantom inventory for extended periods	Identifies phantom inventory cases early
Often increases counting effort	Improves results without increasing counting effort

Why phantom inventory deserves special attention

Not all inventory errors have the same commercial impact.

One of the most costly is phantom inventory.

This occurs when the inventory system believes stock is available, but the shelf is actually empty.

Because the inventory record appears correct, automated replenishment systems do not trigger a replacement order.

The product will remain unavailable until someone physically notices the discrepancy.

The consequences can be significant.

• Customers cannot buy products that appear to be available.
• Store colleagues waste time searching for stock that does not exist.
• Online orders may be cancelled or substituted unnecessarily.
• Sales are lost without anyone realising why.

The research found that predictive stock counting successfully identified more than 80% of these phantom inventory situations before the problem escalated.

That allows retailers to restore product availability much sooner while reducing the hidden costs associated with inaccurate inventory records.

Smarter stock counting, not more stock counting

The research does not suggest that retailers should abandon stock counting.

Regular stocktakes remain essential for financial reporting, compliance and validating inventory records.

Instead, predictive stock counting changes where retailers focus their effort between those major counts.

Rather than applying the same level of attention across every product, retailers can direct colleagues towards the inventory records that data suggests are most likely to be inaccurate.

Stable, predictable products require fewer interventions.

Products affected by promotions, waste, theft or rapid sales patterns receive greater attention.

This approach allows retailers to improve inventory record accuracy without increasing the number of stock counts they perform.

In other words, the objective is not to count less.

It is to count more intelligently.

Can retailers implement predictive stock counting today?

One of the most encouraging findings from the research is that predictive stock counting does not depend on entirely new technology.

The machine learning model was developed using operational data that retailers already collect as part of normal business activities.

That means many organisations already possess the information needed to begin identifying patterns associated with inventory record inaccuracy.

As artificial intelligence and machine learning tools continue to mature, predictive stock counting is likely to become increasingly practical for retailers of all sizes.

Rather than replacing inventory management teams, these systems provide better information to support operational decision-making.

The result is a more proactive approach to inventory management—one that helps colleagues focus on preventing problems rather than simply discovering them.

A new direction for inventory management

Retailers have traditionally viewed stock counting as a measurement exercise.

This research suggests it should become a prediction exercise.

By identifying the inventory records most likely to contain errors before colleagues begin counting, predictive stock counting has the potential to improve inventory accuracy, reduce wasted effort and increase product availability simultaneously.

It represents a shift from reacting to inventory errors towards preventing them.

For retailers facing growing product ranges, increasing labour pressures and rising customer expectations, that could prove to be one of the most important developments in inventory management for many years.

Download the research

Smart Inventory Record Inaccuracy (IRI) Prediction and Management explores how machine learning can identify inventory records most likely to be inaccurate using data retailers already possess.

Drawing on more than 1.3 million stock audits across six grocery retailers, the report explains how predictive stock counting can improve inventory accuracy while making better use of existing resources.

Download the free report today to discover how predictive stock counting could transform inventory management in your organisation.

ECR Retail Loss also runs inventory record accuracy training where the three Professors look at practical implications of their research. You can read a summary of the 2026 IRI Training course here.

Frequently Asked Questions

What is predictive stock counting?

Predictive stock counting uses machine learning to identify inventory records most likely to contain errors, allowing retailers to prioritise stock checks where they will have the greatest impact.

How does predictive stock counting improve inventory accuracy?

Rather than checking every product equally, predictive stock counting focuses on products that historical data suggests are most likely to have inaccurate inventory records. This helps retailers identify more genuine errors while carrying out fewer unnecessary checks.

What is phantom inventory?

Phantom inventory occurs when a retailer’s inventory system shows products as being available even though the shelf is empty. Because the inventory record appears correct, replenishment orders may not be triggered until the error is discovered manually.

Does predictive stock counting replace cycle counting?

No. Predictive stock counting complements traditional cycle counting and statutory stocktakes. It helps retailers decide which products should receive attention between scheduled inventory counts.

Can predictive stock counting work with existing retail data?

Yes. The research demonstrates that predictive stock counting can be developed using operational data that retailers already collect, without requiring entirely new data sources.