Decomposing the Sales Promotion Bump with Store Data

Sales promotions generate substantial short-term sales increases. To determine whether the sales promotion bump is truly beneficial from a managerial perspective, we propose a system of store-level regression models that decomposes the sales promotion bump into three parts: cross-brand effects (secondary demand), cross-period effects (primary demand borrowed from other time periods), and category-expansion effects (remaining primary demand). Across four store-level scanner datasets, we find that each of these three parts contribute about one third on average. One extension we propose is the separation of the category-expansion effect into cross-store and market-expansion effects. Another one is to split the cross-item effect (total across all other items) into cannibalization and between-brand effects. We also allow for a flexible decomposition by allowing all effects to depend on the feature/display support condition and on the magnitude of the price discount. The latter dependence is achieved by local polynomial regression. We find that feature-supported price discounts are strongly associated with cross-period effects while display-only supported price discounts have especially strong category-expansion effects. While the role of the category-expansion effect tends to increase with higher price discounts, the roles of cross-brand and cross-period effects both tend to decrease.