Using segmentation to improve strategy and predictive modelling

Abstract

This paper argues that when segmentation is undertaken before applying a model (any predictive model), this will improve both accuracy and insights. Rather than using a single model that provides average accuracy, segmenting similar behaviours and modelling with potentially different variables and potentially different coefficients drives both accuracy and insights. An example is used with survival modelling for a churn problem in the telecom industry, but the same approach would work for nearly any marketing analytic problem, including product purchases, forecasting, elasticity, lifetime value, same store sales and marcom efficiency.