An important strategic decision within the food industry is to achieve the optimal product portfolio allowing an organization to support its customers in the best possible way. Numerous models exist to categorize products based on their relative metrics like revenue, volume, margin or constructed scores. In order to make a more profound decision whether to keep or to remove a product from the portfolio a closer look into product interdependencies is desirable. Hence, by exploring existing DM-techniques through literature and evaluating those DM-techniques that seem suited in a PPM-context by applying each to a dataset, we aim to identify those techniques that complement a Product Portfolio Management process in the food industry. Three DM-techniques were selected: Dependency Modeling, Change and Deviation Detection, and Classification. Of these three techniques, two were found to be of complementary value in a PPM-context, Dependency modeling and Classification, respectively. Change and deviation detection was found to be of no complementary value in a PPM-context due to it forecasting future data points based on historical data, which results in future data points never exceeding the maximum historical data points. However, change and deviation detection could be of complementary value in another context. Finally, we propose an algorithm to visualize the data-driven product classifications in a standard portfolio matrix which portfolio managers can intuitively understand.
Data mining Food industry Product portfolio management Association rule mining Dependency modeling Change and deviation detection Seasonal time series forecasting Classification Business intelligence