Data Preprocessing and Dataset Construction for Discount Optimization and Demand Prediction

Abstract

We study how trade discounts affect order volume and unit margin for everyday products at Sam Eco Life, a juice and beverage distributor operating in the Samarkand region of Uzbekistan. The central empirical object is a cleaned weekly panel covering 56 delivery regions, 142 products, and twenty-one consecutive weeks (January-May 2026), harmonised into the variable schema of the Dominick's Finer Foods (DFF) scanner database so that one preprocessing and analysis pipeline can be applied to both. We use the DFF bottled-juice category-one of the most extensively studied public retail datasets-as a development benchmark, while the Sam Eco Life panel is the primary object of study. The discount signal is the firm's own revaluation of order amounts, an explicit managerial decision recorded independently of the list price, which makes it a clean treatment variable. Descriptively, discounting is associated with roughly twice the order volume and a correspondingly lower unit margin, quantifying the volume-margin trade-off that drives the firm's pricing decisions. We keep the analysis associational and state its limitations explicitly so that the dataset can be used appropriately in subsequent causal work.