Maximizing on-shelf availability for CPGs with Machine Learning

In this whitepaper, we discuss two problem areas impacting the quality of On-shelf availability

(I) Phantom Inventory - Mismatch in digital vs. physical inventory

This scenario reveals instances of incorrect inventory reporting, wherein the gap between physical and digital inventory files has been creeped in due to some of the reasons mentioned in the previous section. In our analysis, we will be focussing on the phantom inventory scenarios where reported inventory is higher than physical inventory, particularly the negative effects of this on sell-out. This gap is termed here as “phantom inventory”.

Looking at the behavior of Inventory and Sell-out lines in Figure 2 - when the reported inventory level stays at a level >20, we see non-zero sell-out events taking place for 3 days consecutively, indicating that the retailer has “likely” run out of physical stock and has no physical units to present to the prospective customer. These zero sell-out events continue till the store has received more inventory through replenishment, post which sales start picking up again.

(II) Put-away compliance - Delay in shelving the received items

Inventory sent to the store for a special promotion is often marooned in the backroom because store managers didn’t run the promotion or stocked the products in a haphazard way. Per industry best practices, a manufacturer expects a retailer to stock their shelves with the received inventory overnight or within a 24 hours period.

However, looking at the behavior of Inventory and Sell-out lines in Figure 3, we see that as soon as inventory is credited into the store’s digital file, we see zero sell-out events occurring for 3 days consecutively before sales picking up again, “likely” indicating that although retailer has received the scanned inbound stock, the stock is dwelling in the store back-office for 3 days before being put on the shelves.

Notice the word “likely” in both observations above, we call these as likely scenarios because there are always probabilities of false positives, i.e. there could be natural zero sell-out events happening at a given inventory level, and our data points necessarily are not all indicating an inventory compliance issue at retailers location. Example, we may be seeing zero sell-out events at a certain inventory level because there is no customer demand, which in turn could be due to,

Hence, to understand confidence in the Phantom and Put-away non-compliance events, we will substantiate the analysis with demand probabilistic components (taking into account above listed factors) and categorize events into High/Medium/Low confidence buckets.