The presented work covers two tasks, the prediction of policy values up to the maturity of an individual contract and the grouping of contracts, which are combined within our approach. Despite the high importance of grouping in practice, little research on the respective topic is available. We focus on the specific type of term life insurance contract and utilize the concept of neural networks. Our analysis exhibits high levels of accuracy for neural networks predicting policy values. Target values are computed by standard actuarial principles. Moreover, we present a modular approach, how a prediction model for a specific risk feature can be used to group contracts in a supervised way. Our methodology constitutes significant improvements compared to a K-means clustering baseline, of which we optimize the computation of its centroids. Key aspects of our methodology include input scaling, LSTM layers, ensemble technique and nested scaling (alias lambda) layers.