Leveraging Big Data Analytics to Detect Backroom Inaccuracies at Target

In this work we study inventory record inaccuracies (IRIs) at the Target Corporation. We address misplaced backroom inventory, which we refer to as Baffles (Backstock Accidentally Found and Fixed in Location), backroom sublocations recorded as empty or underfull which in reality contain unaccounted for inventory. Baffles are a new type of IRI previously unaddressed in the literature which contribute to sales floor stockouts and reduce backroom storage, adding significant costs to retailers operating at scale. Unlike other forms of IRI such as phantom inventory, resolving Baffles provides not just additional revenue but increased savings: if the misplaced stock is located before it expires or moves out of season, it can be sold to customers without having to order replacement inventory. To detect Baffles we propose a novel machine learning-driven audit recommendation system to predict high-risk backroomshelf sublocations, generating a daily shortlist of audits within backrooms at each of Target's nearly 2000 stores. While similar ideas for recommending phantom inventory audits are wellestablished in the literature and have become increasingly studied in recent years, the nature of Baffles requires new types of signals to address. An analysis of our approach reveals that over the course of 2024 it recommended approximately 17.6 million audits with a success rate of nearly 50 % in recovering unlocated inventory, with 16.3 % of the newly located inventory units resold to customers within 2 weeks of recovery and contributed to improving backroom accuracy by 10.0 %.