Revisiting Optimal Cutoff Score Determination in Direct Marketing Using BLINEX Loss: Bayesian Decisive Prediction Approach

SYNOPTIC ABSTRACTWe develop a method to predict an optimal cutoff point from an ascending list of potential customer activation scores to maximize profit in a future direct marketing campaign. The scoring model is based upon attributes available from a database. Our Bayesian predictive decision theoretic solution is based on two features: (1) a three-parameter bounded, asymmetric loss function BLINEX that has a higher penalty for failure to choose good prospects rather than including too many bad prospects; and (2) a Bayesian nonstandard t-predictive distribution for a future optimal score derived from normal-gamma likelihood function with unknown mean and variance based on data consisting of past optimal activation scores together with a four-parameter conjugate prior. The parameters of the BLINEX loss are also derived from past data. A comparison of BLINEX derived profits with those from traditional squared error loss is made via a fractional factorial experimental design with a half-normal analysis. It shows that our method is far superior based on predicted profit and that, of all parameters, the BLINEX asymmetry and prior mean are the most influential main effects.