Previous Section
I agree with Alan that information collected by retailers via checkout scanners is grossly under-utilized. It is ironic that these data are collected by retailers, but better utilized by manufacturers. Academic research, with a few exceptions, also tends to focus on manufacturer problems. This paper represents a good change of emphasis, towards utilizing checkout scanner data to the benefit of retailers. However, retailers should be more interested in category cross-effects than on the substitution among brands within the same product categories. There is a need to break away from the assumption of weak separability, to consider the substitution or complementarity among different product categories.
Many retail chains also have some form of ``shopping club'' or ``frequent-shopper'' program, which provides them with detailed data on the purchase behavior of its participants, and with the opportunity for micro-marketing at the consumer level. These retailers are able to offer differentiated promotions by direct mail. They could also use the detailed purchase information from the customer database to improve the estimation of the demand systems at the store level. This combination of customer-level ``micro'' data with store-level ``macro'' data could lead to another hierarchical Bayesian model, not that different from the ones discussed in this symposium.
I shall finish my discussion by saying that I have enjoyed reading this well written manuscript. I now have a greater appreciation of the usefulness of hierarchical Bayesian models in solving marketing problems. In the past I had overlooked this approach because of the difficulties in deriving the posteriors analytically. Reading this paper, particularly the two appendices, ( Appendix A: Gibbs Sampler, and Appendix B: Solution of Posterior Distributions) made me realize how feasible, and simpler, the solution can be with the Gibbs sampler.
Go to written version of paper