A SPATIAL CHOICE MODEL FOR PRODUCT RECOMMENDATIONS

Product recommendation models are key tools in customer relationship management (CRM). This study develops a product recommendation model based upon the principle that customer preference similarity stemming from prior response behavior is a key element in predicting current product purchase. The proposed recommendation model is dependent upon two complementary methodologies: joint space mapping (placing customers and products on the same psychological map) and spatial choice modeling (allowing observed choices to be correlated across customers). Using a joint space map based upon past purchase behavior, a predictive model is calibrated in which the probability of product purchase depends upon the customer's relative distance to other customers on the map. An empirical study demonstrates that the proposed approach provides excellent forecasts relative to benchmark models for a customer database provided by an insurance firm. Product recommendation models are key tools in customer relationship management (CRM). An effective recommendation model contributes to the CRM goal of customer expansion by offering high-valued products to regular customers. When regular customers purchase more products from the same company, long-term customer retention is likely to be improved because of increased customers benefits and higher switching costs. This research develops a product recommendation system based upon the principle that customer preference similarity stemming from prior response behavior is a key element in predicting product purchase. Our study brings together two complementary research methodologies: joint space mapping methodology (placing customers and products on the same psychometric map) and spatial choice modeling (allowing observed choices to be correlated across customers). The current research, however, differs from existing research in two respects. First, marketing science models based upon psychometric maps typically examine distances between products and customers in order to infer the attractiveness of each product to each customer (e.g., Moore and Winer 1987). In contrast, this research infers product preferences from the relative similarity of other customers based on inter-customer distances (Kapteyn et al. Because this approach only relies upon customer positions on a psychometric map, it is less susceptible to measurement error and does not require that the researcher accurately locate a new product on an existing product map. Second, contrary to previous work, this research uses a choice model adapted assume that entities (such as customers) can be located in a space. Responses by entities are assumed to be correlated in such a manner that entities near one another in the space generate similar outcomes. The methodology can …