Despite the promise that the technology offers, RFID adoption continues to lag behind initial projections. Industry studies intimate that, in part, this lag is attributable to the RFID performance gap – the difference between desired RFID tag read-rates and those experienced in the field. To explore this phenomena, we employ a mixed research methodology involving intensive case studies of three major retailers as well as large-scale laboratory and field data collection and analysis. Our case studies indicate that major retailers are incurring several operational inefficiencies, including slowed processing times, manual counts to verify inventory levels, and use of “safety factors,” to overcome RFID tag read failures. The practical significance of these findings motivates our subsequent investigation of laboratory test criteria that, when passed, result in RFID tags that perform reliably in the field. To facilitate quantitative analyses, we apply unsupervised learning techniques – i.e., the sequential application of cluster analysis and association rules – to a dataset of 45,416 observations that merge RFID tag laboratory test performance data with read-rate performance data collected from retail supply chains. Our findings identify a pool of RFID tags in which over 99% of the tags have at least a 99% read-rate. Thus, for academics, our study advances a novel unsupervised learning protocol that can be applied to “big data” to gain insights into meaningful supply chain issues, such as RFID tag performance. For practitioners, we establish laboratory test criteria that can be used to identify RFID tags that operate reliably in real-world applications.