Artificially Intelligent Specification and Analysis of Context-Dependent Attribute Preferences

In many markets, customer preferences are context dependent. In the professional marketing literature, this dependence is typically recognized as a “need-state”.Moss and Edmonds (1997) recently reported a model that allows the testing of the qualitative judgements of domain experts in spirits markets against relevant EPOS data of product sales. This paper extends the use of context dependent customer preferences to the case where domain experts lack confidence in their judgements or the judgements are found not to be in accordance with the data. We describe here an algorithm to produce credible alternative models for the domain experts to confirm or develop in light of their wider domain expertise. The algorithm combines random search, genetic programming and evolutionary hill climbing techniques. We report the results of tests using data from markets for alcoholic beverage. The algorithm enabled the largely endogenous production of qualitative descriptions which are both consistent with observed data and deemed credible by domain experts. In a detailed example, the technique is shown to provide extensive insights into the reason for a widely successful brand to have made little impact in one geographically defined market. The algorithm and its implementation are as rigorous and accurate as conventional, purely statistical techniques. They have the additional advantage of cohering with the language of discourse of the domain experts.