Traditional market basket analysis tells us what is (i.e. correlation). To cite a classic, yet most likely an erroneous example, affinity analysis may find that on Thursdays and Fridays customers tend to buy diapers and beer together in a supermarket — i.e. young families stock up for the weekend. The main shortcoming of this information is that, beyond the conclusions that (a) the store should stock up on diapers and beer more heavily on certain days of the week and (b) both products should be placed close to each other in the store, there is little actionable information merchants and marketers can use to make smarter decisions. Affinity analysis tells us nothing about optimal pricing (e.g. should we discount diapers so that customers come in, buy discounted diapers but also grab a pack of beer with a healthy margin?), optimal promotional strategy (e.g. should we place diapers, beer or both in the ad?) or optimal merchandising tactics (e.g. which product should be placed on the endcap?).

Recent advances in the ability to collect and process transaction-level (TLog) data, coupled with exponential increases in computational power and much cheaper storage, all made it possible for grocers, c-stores and other retailers to build in-house or purchase market basket analytic software. Unfortunately, almost all of these solutions deliver non-actionable insights based on affinity analysis. (more…)