Disaggregated Demand Forecasting via Ensemble Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing demand forecasting methods struggle to accurately predict future demand, especially for retailers offering a large number of products, as they fail to effectively account for seasonal changes and are not scalable for complex retail environments.
Innovation Solution
A system and method that utilizes an enterprise forecasting engine to generate aggregate demand forecasts, which are then disaggregated using a computing system with an API and disaggregation service, incorporating ensemble forecasting models weighted based on past performance to capture seasonal effects and provide granular demand predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If simple estimation methods based on past experience are used, then the forecasting process is easy to implement, but the accuracy of demand prediction deteriorates, especially for seasonal products and large retail chains
Solution Approach 1:
The patent segments the forecasting process into multiple components: an ensemble of diverse forecasting models (time series, causal, machine learning), hierarchical aggregation levels (chain-wide, regional, store-level), and separate processing for different product categories. This segmentation allows each component to specialize in specific aspects while maintaining overall system simplicity and accuracy.
Solution Approach 2:
The patent creates a composite forecasting system by combining multiple forecasting models (ARIMA, exponential smoothing, regression, neural networks) into an ensemble approach. Each model contributes different strengths, and they are weighted by historical accuracy to produce a composite forecast that achieves higher precision than any single model while remaining manageable through automation.
2Measurement precision
If statistical models and algorithms relying on past data are used, then the forecasting process becomes more accurate for predictable patterns, but the system fails to capture seasonal changes and complex retail environment variations
Solution Approach 1:
The patent implements dynamic adaptability through models that automatically adjust to changing patterns, seasonal variations, and retail environment changes. The ensemble models continuously learn from new data, and the system dynamically weights models based on their recent performance. Seasonal patterns are captured through time-series decomposition and cyclical feature engineering, allowing the system to adapt to evolving demand characteristics without manual reconfiguration.
Solution Approach 2:
The patent changes parameters and model configurations based on seasonal timing, product category, and retail context. Different models are activated or weighted differently depending on the forecast horizon, seasonality level, and available data. The system adjusts decomposition parameters, seasonal periods, and model selection dynamically to match the specific forecasting requirements of each scenario.
3Adaptability or versatility
If forecasting is performed for a large number of products across multiple locations, then the system becomes more comprehensive, but the computational complexity and processing time increase significantly
Solution Approach 1:
The patent segments the massive forecasting computation into hierarchical levels: chain-wide aggregate forecasts are generated first using simplified models, then disaggregated to regional and store levels using distributional parameters. Product categories are grouped and processed separately, with models selected based on category characteristics. This segmentation reduces the complexity of individual computations while maintaining comprehensive coverage across millions of products and locations.
Solution Approach 2:
The patent applies partial action by generating forecasts at different levels of granularity (aggregate, regional, store, product-level) depending on the specific business question. Not all products and locations require the same level of detail, so the system computes only the necessary level of disaggregation for each query, reducing overall computational burden while maintaining comprehensive capability when needed.
4Productivity
If aggregate demand forecasts are generated for all items within a time period, then the forecasting system becomes more efficient, but the granularity of individual store or date-level predictions is lost
Solution Approach 1:
The patent segments the forecast output into multiple hierarchical levels: aggregate chain-wide forecasts, regional summaries, individual store forecasts, and product-level details. The disaggregation service divides the aggregate forecast into these segments based on requested granularity, allowing efficient aggregate computation while preserving the ability to access detailed store-level or date-level predictions when needed.
Solution Approach 2:
The patent adds a dimension of disaggregation that transforms aggregate forecasts into detailed forecasts by introducing breakdown dimensions such as store location, date, product category, and promotion type. The disaggregation service applies distributional parameters and historical patterns across these dimensions to reconstruct detailed forecasts from aggregate data, effectively adding granularity as an optional dimension without re-computing the entire forecast.
Data Source
AI summary
Methods and systems for forecasting demand for a plurality of items are provided. In particular, the demand forecasting system and methods described herein are useful for predicting demand of products in a retail context. Forecast models are built and used to score incoming sales data to predict future demand for items. Forecast models are validated by evaluating actual demand against predicted demand and using that information to inform how future ensemble forecast will be generated. Forecasts may be broken down into smaller components to satisfy a variety of requests for data from client applications.