Forecasting New Product Sales in a Controlled Test Market Environment

Peter S. Fader
The Wharton School, University of Pennsylvania

Bruce G.S. Hardie
London Business School

Robert Stevens
Information Resources, Inc.

Jim Findley
Information Resources, Inc.

July 2003

Abstract

Although new product forecasting is one of the most critical activities for virtually all firms, it tends to be a source of great frustration and indecision for most of them. To address this need, Information Resources, Inc., set out to create a new forecasting system with particular emphasis on new products in consumer packaged goods markets. We document the resulting modeling system, known commercially as IntroCast, and describe key issues involved in its implementation and managerial interpretations. The model features a "depth-of-repeat" structure, which breaks the new product sales into three underlying components (trial, first repeat, and additional repeat), each of which is modeled independently (including distinct influences for marketing mix effects). We demonstrate the performance of these components -- separately and together -- for a 52-week forecasting horizon and then validate the model on a number of actual new product tests. The model's excellent performance makes it possible to shorten the timeframe required for model calibration from a typical six-month interval down to a twelve-week period, a very significant improvement for today's highly competitive new product marketplace.