Bayesian Multivariate Dynamic Linear Models for SKU Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Retailers face challenges in accurately forecasting sales of stock keeping units (SKUs) due to volatile and incompletely observed demand, leading to issues of overstocking or understocking, which increases holding costs and lost sales opportunities.
Innovation Solution
A system utilizing Bayesian multivariate dynamic linear models to forecast sales by clustering similar SKUs, employing a moving window time series cross-validation scheme to determine training weights for ensemble forecasting, thereby providing more accurate inventory planning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional univariate forecasting methods are used for each SKU individually, then the forecasting process is simple and easy to implement, but the forecasting accuracy is low due to volatile and incompletely observed demand
Solution Approach 1:
The patent combines multiple univariate forecasting models into a single multivariate ensemble model that leverages cross-sectional correlations between SKUs. By grouping related SKUs and modeling their interdependencies, the system achieves higher forecasting accuracy while managing complexity through structured model integration and automated weight optimization.
Solution Approach 2:
The patent creates a universal forecasting framework that can handle diverse SKU types with varying demand patterns. The ensemble model serves multiple functions: it forecasts individual SKUs, captures cross-SKU correlations, adapts to different demand volatilities, and provides robust predictions even with incomplete historical data through shared information across the SKU portfolio.
2Reliability
If more inventory is stocked to prevent stockouts, then the risk of lost sales is reduced, but the holding costs and warehouse space requirements increase
Solution Approach 1:
The patent uses the ensemble forecasting model to predict future demand before the selling period begins, enabling proactive inventory planning. By accurately forecasting which SKUs will be high-sellers, the system allows retailers to pre-position inventory optimally, ensuring stock availability for anticipated high-demand items while avoiding excessive inventory for low-demand items, thus balancing stockout prevention with holding cost reduction.
3Loss of energy
If fewer inventory items are stocked to reduce holding costs, then the warehouse space is optimized, but the risk of stockouts and lost sales increases
Solution Approach 1:
The patent implements a feedback mechanism where the ensemble model continuously learns from historical forecasting performance and actual sales outcomes. The model adjusts the weights of individual univariate models based on their predictive accuracy, allowing the system to adapt to changing demand patterns and improve identification of high-seller SKUs over time, thereby optimizing inventory levels dynamically to maintain stock availability while controlling holding costs.
Data Source
AI summary
A system and method for grouping units for forecasting purposes is presented. A sales forecast for a set of stock keeping units (SKUs) is desired. The SKUs are separated into clusters based on the similarity of the SKUs. Then a set of Bayesian multivariate dynamic linear models is chosen to be used to calculate a sales forecast for each of the clusters of SKUs. The accuracy of each dynamic linear model is determined in a training procedure and a set of weights for each dynamic linear model is calculated. Thereafter, the weights can be used with the dynamic linear models to create a weighted average forecast model. The training procedure can be run periodically to maintain the accuracy of the weights. Each procedure can operate on a sliding window of data. Other embodiments are also disclosed herein.