Abstract. We focus on a real case of the motor insurance sector. We propose four different methods to predict lapsing from a portfolio of policies. We present a comparative analysis between three different performance measures in order to assess the predictive power of each model. Our comparison analyses the outcomes of a logistic regression, a conditional tree, a neural network and a support vector machine. These are all considered basic approaches to data mining and knowledge discovery. The main contribution of this paper is to show that, depending on the type of analysis and the objective of the researcher, the optimal prediction method may differ.

The big-data revolution has impacted the insurance industry more than expected, to become a paradigmatic example of what the new digital economy is. The large amount of data and predictive modeling in insurance represents a turning point and a golden opportunity to channel the theory of risk to the prediction of losses. The changes are radical and demand deep transformations at the organizational level. In this paper we present some reflections on what the incorporation of Analytics implies in an insurance company and we show its inherent complexity through a case of success.