Optimizable and Implementable Aggregate Response Modeling for Marketing Decision Support

The methodological discussion on the calibration of aggregate marketing response models has shifted away from how to obtain usable input for optimization to how to avoid biases in statistical estimation. The goal of this article is to remind researchers that such calibration is performed either to support managers in their marketing-mix decisions, or to create general knowledge that leads to a better understanding of marketing relationships and thus indirectly supports decisions. Both goals require response models that are optimizable. They also must be implementable if actual decision support is aimed at. Herein, I identify several aspects for which these requirements are not always fulfilled: First, the appropriateness of the chosen functional form of the marketing response models is rarely discussed, although different forms imply quite different optimal solutions. Second, endogeneity is taken into account by structural equations, although we lack sufficient information on how managers reach their decisions. Third, estimation methods for response models are often evaluated based on goodness-of-fit, while an assessment of their usefulness for subsequent optimization is neglected. Therefore, I provide recommendations for improving the current practice by better specifying the response function and undertaking more simulation-based evaluations of the best estimation method for use in subsequent optimization. With respect to implementation, usability can be facilitated using spreadsheets and heuristics. Moreover, gaining generalizable and replicable knowledge requires better documentation of results. This is achieved through providing elasticities and as many details as are necessary for replicating a study, thereby enabling faster learning