Predicting Mail-Order Repeat Buying: Which Variables Matter?

In this study, we propose a customer-oriented conceptual model of segmentation variables for mail-order repeat buying behavior. We investigate (1) from a theoretical perspective what customer-related variables should be included in response models for modeling repeat purchasing, and (2) empirically validate how these variables perform for predictive purposes. We use binary logit modeling. Our results confirm that all three traditionally-used R(ecency), F(requency) and M(onetary value) variables are very important in predicting who is going to purchase during the next mailing period, with frequency being the most important one. In total, they account for 50 % of the ‘room for improvement’ in terms of AUC performance. However, next to the RFM variables, our findings suggest that at least three other variables significantly increase the predictive performance of the models: (1) credit usage, (2) length of relationship, and (3) general mail-order buying behavior. Depending on the context of the specific company use of these additional variables may translate into millions Euro of additional profit. Furthermore, we conclude that buying additional data from external sources is not economically justified when predicting repeat purchasing behavior.