On the "Invisible Inventory Conundrum" in RFID-Equipped Supply Chains: A Data Science Approach to Assessing Tag Performance

Recent trade reports suggest that RFID implementation continues to lag lofty projections. A primary concern is that, despite the high cost of implementing RFID systems, realized read-rates fall short of expectations. This results in the invisible inventory conundrum whereby tagged merchandise may still not be accurately represented in inventory records. Drawing from data science to address this issue, we ask: How can directed data mining models be used to identify laboratory test performance criteria for RFID tags that operate reliably across the idiosyncratic facilities (i.e., unique DCs, warehouses, and stores) that comprise apparel retailers' supply chains? We investigate this question by advancing a methodology that integrates laboratory test performance data, field tests of RFID tags fixed to apparel items and scanned under normal operating conditions, and the application of five directed data mining models to the integrated data set of laboratory and field test results. Our analyses of 45,416 observations show that two directed data mining models may identify-with near-100% accuracy-laboratory test criteria that discriminate tags having 99% or greater read-rates in the field. Accordingly, our study validates a generalizable methodology for identifying technical performance standards for tags that operate reliably within apparel retailers' supply chains.