We focus on the multiple roles that browsing behavior plays in assisting our ability to understand and forecast repeat-purchase patterns for an online merchant. Using data from Media Matrix, we decompose buyer behavior into two distinct sub-models: one involving the time between repeat visits to a particular website, and one involving the conversion process through which certain visits are associated with purchase transactions. We show that this integrated two-phase model offers substantially better forecasts and managerial diagnostics than a set of well-known benchmark models that ignore the visit aspects and just model the purchase patterns directly. This is joint work with Wendy W. Moe (University of Texas - Austin).