Retail Sales Forecasting with Promotional Cross-Item Effects
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methodologies for retail sales forecasting fail to accurately account for promotional cross-item (PCI) effects, which can lead to inaccurate sales predictions due to limitations in computational scalability and inability to capture both cannibalization and halo effects, resulting in suboptimal merchandising and revenue management.
Innovation Solution
A computer system aggregates historical sales data to predict PCI effects by forming regression models based on promotional cross-effect attributes, generating predictor variables and model parameters to estimate cannibalization and halo effects, enabling accurate retail sales forecasting.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional sales forecasting methodologies are used, then the forecasting process is simple, but the accuracy of sales predictions is poor due to inability to capture promotional cross-item effects
Solution Approach 1:
The patent segments the forecasting problem by introducing separate promotional cross-item effect variables (PCI effects) as distinct components from baseline sales forecasting. The model divides sales into promoted item sales, non-promoted item sales, and promotional cross-item effects, allowing each component to be modeled and analyzed independently before combining them for the final forecast.
Solution Approach 2:
The patent introduces promotional cross-item effect variables as intermediary components that mediate between the promotion indicators and the final sales forecast. These PCI effect variables capture the indirect effects of promotions on non-promoted items, serving as intermediate calculations that bridge the gap between promotion data and sales outcomes.
2Measurement precision
If computational methods are improved to capture PCI effects, then forecast accuracy increases, but computational scalability decreases
Solution Approach 1:
The patent changes the parameter representation by using aggregated promotional cross-item effect variables that summarize complex interaction patterns into manageable parameters. Instead of modeling every possible item interaction individually, the system aggregates similar PCI effects into unified parameters that can be efficiently computed and scaled across large product catalogs.
3Adaptability or versatility
If the model accounts for both cannibalization and halo effects, then the completeness of sales analysis improves, but the model complexity increases
Solution Approach 1:
The patent creates a universal promotional cross-item effect modeling framework that can simultaneously capture both cannibalization effects (negative PCI effects where promotions reduce non-promoted item sales) and halo effects (positive PCI effects where promotions increase non-promoted item sales). The same general PCI effect variables and regression model structure serve both purposes, eliminating the need for separate specialized models.
Data Source
AI summary
A system that predicts promotional cross item (“PCI”) effects for retail items for a store receives historical sales data for the store and stores the historical sales data in a panel data format. The system then aggregates the stored sales data as a first level of aggregation that is aggregated to the store, a product and a time period. The system further aggregates the first level of aggregation aggregated data as a second level of aggregation that is based on a promotional cross effect attribute (“PCEA”) and is aggregated to the store, the time period and a PCEA level. The system derives PCI effect predictor variables from the second level of aggregation and, for each PCEA within a retail item family, forms a regression model. The system then generates estimated model parameters for one or more PCI effects for each PCEA from the regression models.