ALLOCATING THE DAILY PROMOTIONAL EFFECT BASED ON LOCAL AND HISTORICAL DYNAMICS

Bystrov Aleksandr Igorevich

This article addresses the problem of detailed allocation of the cumulative promotional effect across the days of a promotion in retail trade and the fast-food industry. The aim of the study is to develop a hybrid methodology that combines local dynamics (daily sales forecasts) and historical patterns of promotional effect distribution. The proposed approach is based on calculating a "local" share determined from the sales forecast for each day of the promotion and using an averaged "historical" profile defined by typical patterns for promotions of various durations. The mechanism for blending the two sources of information is implemented through the parameter α, which allows for optimal adjustment of the final sales increment already calculated by global models (e.g., XGBoost). The methodology includes comparing three distribution options—purely local, purely historical, and hybrid—with effectiveness evaluated using the WAPE metric. Experimental results based on real data from a fast-food chain demonstrate that the hybrid approach reduces forecasting error to 24.7%, significantly outperforming alternative methods. The study also emphasizes the potential for integrating the proposed methodology into corporate information systems and using modern machine learning algorithms and neural networks to further improve the accuracy of detailed forecasts.

Keywords:demand forecasting, promotional effect, daily distribution, hybrid profile, local dynamics, historical profile, information systems, machine learning, neural networks