89-6 |
Dynamic prediction models using recursive PLS for increased product success |
L. SAVANT, J. D. Kolsky, D. C. Plaehn, D. S. Lundahl, and G. J. Stucky. CAMO Inc., po box 1628, Corvallis, OR 97339 Consumer wants and needs are ever changing. Therefore, product success requires a development strategy which can adapt in response to evolving needs. An innovative approach to achieve this entails continually updating consumer drivers of liking, whereby product development and marketing teams can better anticipate consumer behavior and react to changing trends in real-time. Traditionally consumer data collection and analysis is conducted as a batch process, with successive data collection cycles requiring larger investments of time and capital. Yet minimal information is available for guidance to product developers in the interim. A solution for continual prediction of consumer trends with simultaneous reductions in cost and time to market is discussed. Recursive Partial Least Squares (PLS) is presented as the tool for this purpose along with a case study. PLS models have been successfully used to determine drivers of liking. However, these models are static - once developed they can only be used to predict new outcomes if the conditions under which the model was built are unchanged. The increased value from recursive PLS models comes from using subsequent data to update the existing model, enabling the initial model to “learn” from the new data in order to provide more accurate predictions. These models can be updated with smaller data increments, thereby saving time and money. An initial data set (calibration set) of consumer product liking scores and analytical data on 7 products from 1998 was used to determine key drivers of liking. Data collected in 2000 was used to update the initial model. The updated drivers of liking using a recursive PLS model are compared to results of treating the data in 1998 and 2000 as ad-hoc studies. The benefit to researchers is quicker and more cost effective access to current consumer information in the face of changing consumer behavior.
Session 89, Product Development
|