Store Assortment Forecasting Using Online-Only Item Demand Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Retailers face challenges in optimizing in-store assortment due to limited shelf space and labor constraints, especially when transitioning from online to physical stores, as existing demand forecasting methods require extensive in-store sales history and struggle with novel ecommerce items lacking such history.
Innovation Solution
A system utilizing a machine learning model trained on shared items sold both online and in-store, leveraging similarity-based demand scores, local popularity, and text similarity to estimate in-store demand for online-only items, enabling assortment optimization based on ecommerce trends.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing demand forecasting methods are used, then in-store demand can be forecasted, but at least two years of in-store sale history is required
Solution Approach 1:
The patent introduces an intermediary approach by using online sales data and item similarity metrics as mediators to bridge the gap between online and offline demand forecasting. The system leverages online demand data for items that are similar to in-store items, using similarity scores and online sales patterns as intermediary information to predict in-store demand without requiring extensive historical in-store sales data.
Solution Approach 2:
The patent applies preliminary action by pre-calculating and storing item similarity scores, category relationships, and online demand patterns before the actual demand forecasting is needed. This preliminary data preparation allows the system to quickly estimate in-store demand for new items without requiring years of historical data collection, as the foundational similarity metrics and patterns are already established.
2Adaptability or versatility
If novel ecommerce items are introduced to stores, then assortment freshness and relevance can be improved, but lack of in-store sales history makes demand prediction difficult
Solution Approach 1:
The patent uses copying by creating virtual copies of in-store item demand patterns through online data. Instead of requiring actual in-store sales history for novel items, the system copies demand prediction patterns from similar items that do have in-store sales data. By replicating the demand forecasting approach using analogous items with known performance, the system can accurately predict demand for new items without their own historical data.
Solution Approach 2:
The patent treats historical sales data as a disposable resource that should be utilized efficiently. Rather than requiring long-term retention and analysis of extensive historical in-store data, the system uses readily available online data and similarity metrics as temporary, easily obtainable substitutes that provide sufficient predictive power for demand forecasting of novel items.
3Productivity
If shelf space is allocated to novel items, then customer demand can be captured, but labor cost for replenishment and supply chain management increases
Solution Approach 1:
The patent implements feedback mechanisms by continuously monitoring actual in-store sales performance of novel items against predicted demand and using this feedback to refine future predictions. The system learns from actual customer behavior and sales data, adjusting demand forecasts and assortment recommendations accordingly. This feedback loop enables the system to optimize shelf space allocation and supply chain management by adapting to real-world performance patterns.
Data Source
AI summary
Systems and methods for estimating in-store demand based on online data are disclosed. In some embodiments, a machine learning model is trained based on data of shared items that are being sold both online and in-store by a retailer. For a physical store of the retailer, inference items are determined from items being sold online but not in-store. An estimated demand is computed for each inference item to be offered for sale in the physical store in a future time period, based on the trained machine learning model and online data of the inference item. Based on the estimated demands for the inference items, recommended assortment data is generated for the physical store in the future time period, and is transmitted to a computing device associated with the physical store for assortment refresh.