One of the key challenges in managing the launch of a new product is the attempt to obtain valid indicators about the product's likely future sales levels before the product is actually launched. Previous methods have focused on managerial judgment, consumer surveys, and other processes that solicit opinions from which a forecast is generated. In this paper, we develop a model that generates forecasts based on behavioral data, namely the pattern of advance orders that occur prior to the actual launch and release of the product.
We present a duration model that incorporates the underlying concepts of new product diffusion in a Bayesian framework. Specifically, we model product sales as a mixture of two Weibull distributions, one representing the behavior of "nnovators," and the other representing the "followers." We utilize hierarchical Bayes estimation to allow the parameters of these segments to be correlated for each new product and also to accommodate heterogeneity across products.
The model is applied to 66 music CDs (obtained from CDNOW). We demonstrate the diagnostic insights provided by the model, as well as its excellent forecasting ability.
|