Given the emerging concept of a customer-centric approach to marketing, customer relationship management (CRM) has seen increased attention. Among essential tools to implement CRM is customer base analysis which seeks to understand and predict transaction patterns of individual customers. This dissertation, composed of three essays, studies the dynamics of shopping behavior in customer base analysis and its implications for CRM. The first essay provides an overview of modeling approaches for customer base analysis, reviews relevant research in the marketing literature, and identifies an agenda of areas that are in need for further research. The second essay proposes a modeling framework for multi-category customer lifetime value (CLV) analysis in a non-contractual setting. To this end, we model customers' arrival process, purchase incidence and amount decisions across categories, and latent defection in an integrated framework. The proposed framework makes use of a latent space model that parsimoniously captures various dynamics of multi-category shopping behavior arising from the interplay between purchase timing and choice across categories. Using category-level transaction data from a leading beauty care company, we show that the proposed model offers excellent fit and performance in predicting customer purchase patterns across categories. Our model allows one to quantify the contribution of individual categories to CLV and assess the relationship between shopping basket choice and CLV. The third essay examines shopping behavior of online customers. We develop a model that captures the clustered visit patterns of online customers and predicts how a series of store visits lead to a purchase. Our model is based on the notion that the arrival process of customer visits consists of multiple visit clusters with relatively short intervisit times within a cluster and a longer intervisit time between clusters. Because the start and the end of each visit cluster are unobserved, we employ a changepoint modeling framework and statistically infer the cluster formation on the basis of customer visit patterns through data augmentation in Bayesian approach. In our empirical analysis using data from a major e-commerce site, we find strong empirical evidence of lumpy shopping patterns by online customers with significant heterogeneity in the extent of the lumpiness. As part of our substantive contribution, we show that taking into account the clustered visit patterns can significantly improve the model performance in predicting purchase conversions across store visits.