Fast-Track: Article Using Advance Purchase Orders to Forecast New Product Sales

Marketers have long struggled with developing forecasts for new products before their launch. We focus on one data source—advance purchase orders—that has been available to retailers for many years but has rarely been tied together with postlaunch sales data. We put forth a duration model that incorporates the basic concepts of new product diffusion, using a mixture of two distributions: one representing the behavior of innovators (i.e., those who place advance orders) and one representing the behavior of followers (i.e., those who wait for the mass market to emerge). The resulting mixed-Weibull model specification can accommodate a wide variety of possible sales patterns. This flexibility is what makes the model well-suited for an experiential product category (e.g., movies, music, etc.) in which we frequently observe very different sales diffusion patterns, ranging from a rapid exponential decline (which is most typical) to a gradual buildup characteristic of “sleeper” products. We incorporate product-specific covariates and use hierarchical Bayes methods to link the two customer segments together while accommodating heterogeneity across products. We find that this model fits a variety of sales patterns far better than do a pair of benchmark models. More importantly, we demonstrate the ability to forecast new album sales before the actual launch of the album, based only on the pattern of advance orders.