A neural network application to consumer classification to improve the timing of direct marketing activities

Abstract

This article develops an alternative estimation approach for classifying new prospective consumers as “good'' or “bad'' prospects for direct marketing purposes. We show that the traditional approach of using demographics alone to profile non-active consumers (those who have yet to buy in the category) can be improved by waiting to observe their initial and limited number of sequential purchases in the category. We call this method the early purchase classification (EPC) approach. We make use of two established classification models, a multinomial logit model (MNL) and a neural network model (NN), and show that the classification accuracy of both models using our EPC approach outperforms the traditional approach of classifying non-active prospects using demographics only. Furthermore, we find that the NN model consistently outperforms the MNL model at this task. This research uses the best aspects of each model by utilizing the MNL model to determine which variables are most relevant to the classification and then using those variables for classification in the NN model. Using the complementary features of the MNL and NN models, managers can use the EPC approach to determine the most profitable time in a purchasing history to classify and target prospective consumers new to their categories.