Store and DC Item Ranging Using ML Demand Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional supply chain models face challenges in matching supply with demand, leading to issues such as excess or lack of inventory, due to factors like seasonality, weather, location, and demographics, especially in complex ecosystems with interconnected stores and e-commerce channels, necessitating improved methods for sizing item ranges in distribution centers and stores.
Innovation Solution
A processor-implemented method using supervised machine learning (ML) and reinforcement learning (RL) models to determine the range of items based on sales data, geography, demography, and affinity, pre-processing the data to eliminate skewed information and creating categorical sales bins, and implementing AI models for inventory forecasting and range determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional explicit rules and engineering systems are used to manage inventory, then the system is simple to implement, but it leads to excess or lack of inventory due to inability to adapt to varying factors
Solution Approach 1:
The system enables self-service through automated machine learning models that independently analyze sales data, seasonality, weather, location, and demographic factors to generate inventory recommendations without requiring manual rule configuration. The models continuously learn and adapt to changing conditions, performing the complex analysis work autonomously.
Solution Approach 2:
The patent replaces manual rule-based engineering systems with automated machine learning algorithms. Instead of mechanically applying predefined rules, the system uses supervised and reinforcement learning models to dynamically determine optimal inventory levels based on multiple varying factors, substituting mechanical decision-making with intelligent computational approaches.
2Measurement precision
If manual rule-based systems are used for inventory management, then the implementation is straightforward, but the accuracy of matching supply with demand is low
Solution Approach 1:
The system performs preliminary action by pre-processing sales data and training machine learning models in advance. The models are trained on historical data containing seasonality, weather, location, and demographic information before being deployed to make real-time inventory recommendations, allowing the system to quickly provide accurate predictions without processing everything from scratch.
Solution Approach 2:
The patent uses copying by creating simplified categorical representations of complex data patterns. The machine learning models learn to copy the relationships between various factors (seasonality, weather, location, demographics) and inventory outcomes from historical data, then apply these copied patterns to new situations for rapid and accurate decision-making.
3Measurement precision
If the supply chain system processes all available data without pre-processing, then all information is considered, but skewed data leads to inaccurate predictions
Solution Approach 1:
The system applies taking out by extracting and removing skewed or problematic data points during the pre-processing stage. The machine learning pipeline includes steps to identify and eliminate biased or erroneous data before training, ensuring that only clean, representative data influences the model predictions while maintaining overall system simplicity.
Solution Approach 2:
The patent uses parameter changes by transforming raw data into optimized formats suitable for machine learning. This includes normalizing numerical values, encoding categorical variables, and adjusting data distributions to eliminate skewness. These parameter transformations improve prediction accuracy while providing a structured approach to data processing.
4Loss of energy
If conventional models are used, then operational costs are higher due to excess or lack of inventory, but implementing advanced AI models increases initial system complexity
Solution Approach 1:
The system implements feedback through reinforcement learning mechanisms where the model continuously receives performance feedback from actual inventory outcomes. The system learns from the consequences of its recommendations, adjusting its strategy to minimize excess and shortage inventory over time. This automated feedback loop reduces operational costs while the system manages its own complexity through self-learning.
Data Source
AI summary
Conventional techniques implemented various explicit rules and engineer systems to deliver a solution. Stores and Distribution Centers (DC's) are faced with problems related to excess inventory/lack of inventory which vary with several, and the biggest challenge lies in matching supply with demand. Present disclosure provides systems and methods that perform node ranging wherein range of items are stocked in stores and DCs for fulfilling demand. The system identifies stores and DCs and further recommend store specific items based on a request received by the system and then implements a supervised machine learning model and a reinforcement learning models determine a range of items to be sized from the recommended store specific items which are based on most frequent line items being identified from a set of inputs received therein.