Scalable Assessment of Cross Category Promotion Effects Using Machine Learning

Direct effects of promotions are studied extensively in marketing research although indirect effects (cross category research) are less considered. Indirect effects, cannibalization and halo, are typically studied in highly stylized setups (e.g. pancake and syrup) or in setups that involve a handful of categories. Parametric models, mostly regression models with heterogeneity, are usually used to study the indirect effects of promotions on pre-specified items. However, from retailers perspective that cares a great deal about indirect effects the retailing context involves a very large number of categories. Therefore, existing models are not scalable to study indirect effects of promotions for a retailer. In this paper, we propose a framework that is designed to uncover the magnitude and direction of indirect effects. The conceptual framework relies on behavior based Jaccardian index of similarity to assess indirect effects. Machine learning tools are used to uncover items that are independent, exhibit cannibalization or halo effects. The conceptual framework is simple to use, offers usable insights for a retailer and is scalable.