Stitch Fix has been granted a patent for a system that uses data about clients and items to recommend items from an inventory. The system includes feature selection processes and recommendation processes that are modified based on feedback from clients and selection entities. The modification of these processes helps in improving the selection of items for clients. The system also implements machine learning techniques for recommendation purposes. GlobalData’s report on Stitch Fix gives a 360-degree view of the company including its patenting strategy. Buy the report here.
According to GlobalData’s company profile on Stitch Fix, AI assisted CAD was a key innovation area identified from patents. Stitch Fix's grant share as of September 2023 was 49%. Grant share is based on the ratio of number of grants to total number of patents.
Recommendation system for selecting items based on client feedback
A recently granted patent (Publication Number: US11775929B2) describes a system that utilizes machine learning techniques to recommend items to clients. The system includes one or more data stores that store client attribute data and client history data for multiple clients, as well as item attribute data for items in an inventory. The system also includes processors that select item attributes and client specified attributes for presentation to an entity responsible for selecting items for clients. The selection is made using feature selection processes.
The system further selects a subset of recommendation processes to use for each client and identifies different subsets of recommended items using these processes. The recommendations are provided to the entity responsible for selection, and feedback is received from both the entity and the client. The client feedback is used to update the client history data, and the feature selection processes are modified based on this feedback. This modification includes adjusting the weight associated with the selected client specified attribute.
Additionally, the system can systematically change the features selected for presentation and identify which features result in positive or negative client feedback. Based on this information, the feature selection process is modified to prioritize features that lead to positive feedback and minimize the presentation of features that result in negative feedback.
The patent also mentions that each recommendation process utilizes the client attribute data, client history data, and item attribute data in different ways. The system can perform simulations based on the client history data to modify how client attributes or item attributes are displayed.
Furthermore, the system can access the client history data to generate training tasks for entities responsible for selecting items. These tasks are presented to the entities, and training feedback is provided based on the client history data. The system can also store entity attribute data, including training history data, and assign tasks to entities based on this data.
Overall, this patented system utilizes machine learning techniques, client feedback, and feature selection processes to recommend items to clients and improve the performance of the recommendation system based on client history data.