Consumers dynamically update their preferences over time based on information learned through product search and consumption experiences, particularly in online media. Using three unique datasets from different domains, we address specific ways in which firms can use rich information about their customers' behaviors to improve (1) the visual display of products on a webpage in online shopping, (2) predictions of new product adoption in online gaming, and (3) the timing of product release in online learning. First, we explore how consumers visually search through product options using eye-tracking data from two experiments conducted on the websites of two online clothing stores, which can inform retailers on how to position products on a virtual webpage. Second, we examine how consumers' variety-seeking preferences change depending on past consumption outcomes within the context of an online multi-player video game, which can be used to improve predictions of new product adoption. Third, we use clickstream data from an online education platform to test theories of goal progress, knowledge accumulation, and boundedly rational forward-looking behavior, which can be used to explain binge consumption patterns and inform content providers on the best way to structure and release content. In each of these three projects, we build a mathematical model of individual decisions, with the parameterization grounded in theories of consumer behavior, and we demonstrate through in-sample prediction that our model is able to capture specific heterogeneous patterns within the data. We then test that our model is able to make out-of-sample predictions related to managerial interventions, and empirically verify our predictions using either lab experiments or new field data following a natural experiment policy change.