FMCG Retail and Unknown Stockouts

We all know how frustrating it is to check company’s website and verify that they have inventory for the product you’re looking for and then drive to the store just to find out that they don’t have it - you just became another victim of “phantom inventory”. Stockouts cost retailers around 4% of revenue in the United States and closer to 10% in Europe! Revenue loss is not the only challenge as consumers spend on average 21% more time at the store when there’s a case of stockout particularly when it’s due to phantom inventory as the staff end in the “the system tells we should have it in the store, let’s go look for it” cycle.

There are many reasons to stockout and here’s a great research paper done by Thomas W. Gruen, Ph.D. from University of Colorado at Colorado Springs, USA And Dr. Daniel Corsten from IE Business School Madrid. That study is very comprehensive and we recommend checking it out. In this shorter blog, we’ll focus on the “unknown” stockout that we also refer to “virtual” stockout.

So what is a virtual stockout? Essentially, it’s a stockout caused by phantom product or item inventory data in the store system. In essence, the system has positive inventory that in reality doesn’t exist in the store. This is quite understandable as the product level inventory data in the systems is often inaccurate. The before mentioned study showed one retail chain having only 41% of SKUs inventory data being completely accurate and and nearly 20% of SKUs having ±5 or more discrepancy. Traditional stockout is a situation where the store has no sales for an item and inventory shows 0. This is very easy to understand and manage from replenishment perspective. It gets a lot more complicated when the system expects stores to have inventory when in reality the inventory level is lower and potentially even zero. Without the signal from the POS, the fulfillment system will not replenish the store.

So why does this happen? There are many reasons for inventory levels to drop such as theft, breakage, misplaced inventory etc. But there are also supply chain issues such as distribution center picks the wrong products or trucks deliver the inventory to the wrong store. Due to the fast moving pace of retail, it’s more a rarity than norm that all deliveries to store are checked and inventoried. In some cases, palettes and cases have RFID but it’s mostly done with items with higher value.

How do we address this?

There are several ways to mitigate the inventory data quality issues. Store could do more frequent cycle counts but if they don’t know where to focus, it can become incredibly expensive from staff time perspective and the outcome of manual inventory counts are rarely much more than 85% accurate. Retail chain could introduce RFID to all the items but this could become cost prohibitive as stickers can cost 2-4 cents each and stores would need either manual readers or ceiling/shelf installed physical devices. Finally, you could take a data driven approach. We’ll focus on the data driven approaches in the following comparing statistical models to machine learning models.

Statistical modeling

This is the more traditional approach trying to assess phantom inventory issue using a corporate data lake or data warehouse and using statistical tools and visualization with PowerBI or Tableau etc. The challenge with this approach is that we’re looking backwards and will have to account demand shifts, promotions, seasonality and other variables leading to a lag time of 4-8 weeks. Having a virtual stockout unnoticed for 4-8 weeks can result in significant revenue loss especially when the item count for virtual stockout can be upwards to 25% of all SKUs in the store.

Machine learning modeling

ML models use similar statistical approach but due to large dataset and ability to learn from past data, the models will develop patterns and recognize scenarios looking forward where there’s a likelihood for an item having virtual stockout. The Kloud9 model has been tested with audited data reaching over 97% accuracy in predicting a virtual stockout (predicted inventory is zero when system shows positive levels). Being able to identify the store/SKU combinations with virtual stockouts immediately allows for the stores and supply chain/logistics teams to react faster and not only replenish the stores but also adjust inventory systems to improve the data quality over time. Find out more about our solution here.

‍

Why should a retailer care about this?

Addressing unknown stockout / virtual stockout will help retailers capture otherwise lost revenue. Depending on the size and type of the retailer the revenue capture can climb to tens of millions of dollars annually. Obviously knowing that there’s an issue is only a start. The retailer staff will need to take actions to mitigate the problem and for that Kloud9 solution offers also a graphical interface letting users prioritize actions such as verifying current inventory, placing emergency orders, looking for the items within the store etc. It’s very possible that the patterns identify repeating issues with the supply chain that helps the staff to focus on potential root causes with store operations or issues with logistics from DC to store.

‍