Abstract
Companies need tools and models that accurately describe and forecast the diffusion process of new products. Most tools for predicting the diffusion of an innovation, given early data, use aggregate models (such as the Bass model) that capture the impact of advertising and word-of-mouth on adoption. Because of their aggregate nature, however, (i.e., these models treat each month or each year as one observation), they are often unable to produce reliable predictions early in the diffusion process, when such predictions would be most helpful (e.g., in order to adjust the launch campaign). Moreover, model assumptions about the influence of social interactions on individual consumers are often not well defined, and it is often unclear when a given model is appropriate.
This study addresses these limitations by combining the traditional aggregate approach with the agent-based approach. Toubia, Goldenberg, and Garcia's proposed models are based on explicit assumptions about the adoption process, and model parameters are expressed as functions of well-defined, measurable variables describing individual consumer behavior and network interactions.
They demonstrate how their approach enabled a major consumer packaged goods (CPG) manufacturer to develop a theory-driven aggregate diffusion model and to calibrate this model shortly after launch using survey data.
The proposed approach addresses the limitations of aggregate diffusion models on three fronts. First, it allows developing rich models that are based on explicit assumptions about the adoption process. Second, because the parameters of the proposed models are linked to well-defined, measurable variables describing individual consumer behavior and network interactions, these models may be calibrated using survey data shortly after launch. Third, because they are based on a set of explicit assumptions, the authors are able to provide guidelines to help managers and researchers identify conditions under which they provide an accurate description of the diffusion process, and conditions under which they should be replaced with alternative diffusion models.