Machine-Learned Bulk Order Segmentation Across Retail Sources
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional fulfillment models fail to effectively leverage available resources to efficiently cater to bulk orders, particularly in business-to-business scenarios, lacking the ability to predict conditions such as travel times and source selection, leading to inefficient inventory management.
Innovation Solution
A machine learned model is employed to determine segmenting options for fulfilling bulk orders by analyzing availability information from various sources, including consumer packaged goods (CPG) warehouses and retailers, to identify optimal combinations of pickers and sources for item fulfillment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional fulfillment models are used, then the system is simple to operate, but it cannot effectively leverage available resources to fulfill bulk orders efficiently
Solution Approach 1:
The fulfillment system segments bulk orders into multiple smaller orders that can be fulfilled by different retailers. The machine learned model divides the total quantity request into segments, assigning portions to different retailers based on their availability and capabilities, thereby enabling efficient bulk order fulfillment without requiring a single complex retailer to handle the entire order.
Solution Approach 2:
The machine learned model acts as an intermediary between the user's bulk order request and the multiple retailer sources. It processes the request, predicts fulfillment conditions, and determines optimal segmenting options, mediating the complexity of coordinating multiple sources while presenting simplified solutions to the user.
2Reliability
If conventional fulfillment models are used, then the system is easy to understand, but it cannot predict conditions such as travel times and source selection
Solution Approach 1:
The machine learned model performs preliminary prediction of fulfillment conditions, including travel times and source selection, before the actual fulfillment process begins. By predicting these conditions in advance based on historical data and patterns, the system can provide accurate estimates and optimize the fulfillment plan before execution.
Solution Approach 2:
The system incorporates feedback loops where actual fulfillment data is used to train and refine the machine learned model. This feedback mechanism continuously improves prediction accuracy by learning from real-world outcomes, allowing the model to better predict travel times, availability, and other fulfillment conditions for future orders.
3Quantity of substance
If a single retailer is used for bulk orders, then the fulfillment process is simple, but it cannot cater to large item quantities
Solution Approach 1:
The system automatically segments large quantity requests across multiple retailers, with the machine learned model determining the optimal distribution. This segmentation enables fulfillment of bulk quantities that exceed the capacity of any single retailer while maintaining ease of operation through automated decision-making and user-friendly interfaces.
Solution Approach 2:
The machine learned model provides universal functionality by handling both single-retailer and multi-retailer scenarios through a unified system. It adapts to different order sizes and automatically determines the optimal fulfillment strategy, whether consolidating with one retailer or distributing across multiple sources, thereby maintaining operational simplicity across all cases.
Data Source
AI summary
An online concierge system (“the system”) determines that a shopping list from a user client device includes a request for a quantity of an item that exceeds a quantity that can be fulfilled using a single retailer. Responsive to the determination, the system retrieves model inputs based in part on the request. The system determines segmenting options, for fulfilling the request using multiple sources, and their associated costs using a machine learned model and the model inputs. The segmenting options include different combinations of pickers and sources that can be used to fulfill the request. The system provides one or more of the segmenting options and their associated costs to the user client device. Responsive to receiving, from the user client device, a segmenting option of the one or more segmenting options, the system fulfills the request in accordance with the segmenting option.