Compartmental diffusion modeling: Describing customer heterogeneity & communication network to support decisions for new product introductions

Following the introduction of the well-known Bass model and its variations to model adoption of a new product, compartmental modeling has emerged, in which the adoption pattern is modeled at the level of each compartment (or customer segment). Compartmental models add spatial and psychographic segments to Bass-type models which can allow modelers to approximate the heterogeneity, clustering and communication network of customers, while retaining an analytical structure. These models have been mainly explored in the context of demand forecasting and describing customer behavior, but their suitability as a scalable modeling tool to support large scale marketing and/or operational decision making is not explored. In this paper, we propose a flexible compartmental model and assess its suitability in terms of use of data, adherence to micro-level customer behavior, and use in large scale decision making. We show the merits of the CDE model by carrying out an extensive simulation study and also estimation on data on boradband adoption. We find that compartmental models result in estimates that are drastically less biased and can predict the shape of the adoption curve significantly better than what can be achieved by the Bass model. Even though these models can be scalable to capture large number of segments, we show that these improvements can still be observed with even a small number of segments (in contrast to what has been hypothesized in the literature), and are not sensitive to errors in the underlying process of customer segmentation. Therefore, the analytical structure of compartmental models can be further explored to further support large scale marketing and operational decision making, as they do not inherit the major shortcomings of Bass-type models while they capture the effect of customer clustering and interactions. Even though the focus of this paper is on new product adoption, compartmental modeling can also be successfully utilized in diverse physical, biological, and social settings that are governed by similar dynamics.