The business relevance of customer churn analysis is increasing due to the growing availability of corresponding data and intensifying competition. Here, especially the predictive accuracy of modeling approaches is in the focus of researchers and practitioners alike, with deep neural networks recently becoming an attractive method due to their high performance in a variety of fields. However, from a practical point of view, other factors such as the ease of application and model interpretability are also to be considered. These aspects are generally viewed as shortcomings of deep neural networks. Therefore, a novel framework for the application of deep learning in churn analysis is developed and tested in a practical setting. It is shown, that a less complex application procedure and more easily interpretable prediction modeling can be achieved.