Demand Prediction Using Sparse Transaction Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Businesses face inefficiencies in determining which retail stores to visit and when, as existing methods rely on institutional knowledge or static schedules, leading to unproductive visits due to sparse historical sales data and lack of consideration for current promotions and business factors.
Innovation Solution
A computer-implemented system predicts demand by using a two-stage pipeline that includes a buy/no-buy model and a quantity forecasting model, trained with historical transaction data and promotion information, to generate predictions for transaction dates and quantities, optimizing salesperson routes and schedules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If salespeople use static visit schedules or institutional knowledge to determine which retail stores to visit, then they can maintain simple operational procedures, but they experience reduced productivity due to unproductive visits
Solution Approach 1:
The system enables self-service by automatically generating visit schedules and predicting transaction quantities without requiring manual analysis by salespeople. The machine learning models autonomously process historical transaction data and promotion information to produce optimized visit recommendations, freeing salespeople from manual planning while improving visit productivity.
Solution Approach 2:
The patent replaces the mechanical system of manual visit scheduling based on institutional knowledge with an automated computational system. Machine learning models analyze transaction data and promotion information to generate predictions, substituting human intuition and manual processes with algorithmic decision-making that improves visit effectiveness.
2Measurement precision
If businesses rely on sparse historical sales data alone to predict demand, then they can maintain simple data collection processes, but they suffer from measurement precision deficiencies in demand forecasting
Solution Approach 1:
The system merges multiple data sources including historical transaction data, promotion information, and hierarchical category structures to compensate for sparsity in individual datasets. By combining these diverse information sources and processing them through machine learning models, the system achieves accurate demand predictions even when historical sales data alone is insufficient.
Solution Approach 2:
The patent implements a multi-functional prediction system that handles multiple hierarchy levels (from individual items to categories to stores) and various data types (transaction data, promotion data, temporal patterns). This universal approach allows the same system to accurately predict demand across different granularities and contexts, improving measurement precision without requiring separate systems for each scenario.
3Loss of time
If salespeople visit all retail stores on fixed schedules, then they can ensure comprehensive market coverage, but they waste time on unproductive visits where no transactions occur
Solution Approach 1:
The system transforms static visit schedules into dynamic, adaptive plans. Machine learning models continuously analyze current promotion information and historical patterns to generate time-sensitive predictions about which stores are most likely to transact. This dynamic approach allows visit schedules to adapt to changing conditions, reducing time wasted on unproductive visits while maintaining necessary market coverage.
Solution Approach 2:
The system performs preliminary analysis of transaction probabilities before salespeople depart for visits. By predicting which stores are most likely to result in productive transactions based on historical data and current promotions, the system prepares optimized visit sequences in advance. This preliminary action ensures salespeople prioritize high-probability targets, minimizing time spent on unproductive visits.
4Measurement precision
If businesses use detailed hierarchical category structures for organizing products, then they can improve prediction accuracy across different product levels, but they increase the complexity of data processing
Solution Approach 1:
The system segments the prediction problem into multiple hierarchy levels (individual items, product categories, and store levels). Machine learning models are trained separately for each level, allowing specialized processing for different granularities. This segmentation enables accurate predictions at each level while managing complexity through modular model architecture and separate training datasets for each hierarchy tier.
Data Source
AI summary
The present disclosure involves systems, software, and computer implemented methods for proactively predicting demand based on sparse transaction data. One example method includes receiving a request to predict transaction quantities for a plurality of transaction entities for a future time period. Historical transaction data for the transaction entities is identified for a plurality of categories of transacted items. The plurality of categories are organized using a hierarchy of levels. Multiple levels of the hierarchy are iterated over starting at a lowest level. For each current level in the iteration, features to include in a quantity forecasting model for the current level are identified. The quantity forecasting model is trained using the identified features.Predicted transaction dates are predicted for the current level by a transaction date prediction model. The quantity forecasting model is used to generate predicted quantity information for the current level for the predicted transaction dates.