Central to both monitoring and forecasting the performance of a new product is the decomposition of total sales into its trial (i.e., first-purchase) and repeat (or replacement) purchase components. Several researchers have developed models of new product sales that specify submodels for trial and repeat sales, yet can be calibrated using only aggregate sales data (as opposed to data on the underlying sales components). Some researchers have used these models for forecasting purposes, while other researchers have used these models to make inferences about the underlying trial and repeat components of new product sales in situations where only the aggregate sales data are available.
In this paper, we demonstrate that extreme caution must be used when trying to make such inferences about the underlying trial and repeat components of a new product's total sales. Using panel data for twenty new products, we aggregate the household-level transaction data to arrive at aggregate sales data. We fit a model of new product sales to these data and compare the implied trial/repeat patterns to the actual patterns observed using the raw panel data. Looking at several different underlying model specifications, we find that the implied trial/repeat patterns to do not reflect the true patterns. Thus any inferences derived from the aggregate sales data using such models can be very misleading.