A data driven framework for early prediction of customer response to promotions

The study proposes a framework based on functional analysis of transaction data to predict customer spending during promotional events. Retailers face the challenge of considering an extant number of variables and of accounting for methodological constraints while modeling customer response to promotions. Noise and uneven distribution of spending further introduces error into traditional models’ estimates. We represent each customer’s spending as a continuous curve that accounts for heterogeneity due to the purchase cycle as well as cross sectional differences. In order to obtain an optimal functional representation we utilize a data driven iterative procedure. Statistical information from the collection of spending curves is drawn using functional data analysis (FDA). Analysis of a real customer dataset from a North American retail chain shows that the dynamics of information captured by optimal functional representation of transaction data significantly improves predictions for out of sample observations. Framework provides a guideline for modeling fast moving, heterogeneous and noisy customer related data to assist managerial decision making during specialized events such as promotions.