As supermarkets have developed the capability to gather customer data, their focus on customer loyalty has increased; however, retailers cannot actually measure loyalty. In this paper, we propose and test a hierarchical Bayesian approach to predicting customers' share-of-wallet loyalty that offers a high degree of predictive accuracy and discriminates well between loyal and non-loyal customers. Among the types of information that retailers may gather, we find that geographic information (travel time to the store and retail concentration around that store) makes the greatest predictive contribution. Finally, we show how retailers can use the posterior distribution of customer share-of-wallet to more profitably select customers to receive targeted marketing offers.