Understanding customer behavior is becoming critical for electric grid operations. Grid operators need to model customer behavior from multiple perspectives, in part due to recent changes of the customer role from passive loads to prosumers. Customers are active agents rather than 'passive loads,' and this change in behavior comes with a variety of challenges. Solving these challenges requires a 360-degree view of the customer, which calls for machine learning techniques for classification, time-series analysis, and uncertainty quantification. This allows utilities to actively work in the challenging landscape of active customer behavior. In this paper, we describe several important techniques for practical modeling of active customer behavior, together with corresponding areas of behavior modeling: energy savings potential, adoption of sustainable products and services, prediction of photovoltaic adoption, and fraud detection. For each of these applications, we briefly survey machine learning tools that have allowed demonstrable practical impact. Where possible, we illustrate results using datasets from Alliander N.V. (a Dutch energy company), as well as other companies. We present quantitative results, describe their qualitative impact on the companies concerned, and provide some practical insight into model building and validation.