Technology-based piloting for the automated assignment of products to the category tree

Case Study

Background

Rakuten`s German online marketplace is faced with the problem of a very messy category tree in which many of the products are sorted incorrectly. This results from the fact that the manual categorization of products in the category tree by the seller as part of the listing process has a considerable margin of error. A random sample analysis identified up to 50% incorrectly categorized products from a total range of over 10 million items. This incorrect categorization of around 5 million products has a considerable impact on the findability of the products, both onsite and offsite, which leads to significant losses of sales.

Goals

The primary goal of the project is the development and implementation of a technology-based approach for the automatic and correct categorization of the category tree. The existing category tree is to be tidied up and newly listed products are to be automatically sorted correctly. This approach is intended to create a system that reduces the manual subsequent work of checking and tidying up and increases accuracy. The long-term goal is a permanently tidy category tree that also correctly records newly listed items without changing sales processes of the sellers.

Solution

The basic idea for the solution is based on the idea that category trees record and structure products by similarity. The technology for the solution therefore relies on product vectors to capture and compare product descriptions in a technological and automated way. A configurable system makes it possible to define the desired degree of similarity of products at category tree level, enabling iterative improvements. The first prototype version has already reduced incorrect categorization by over 50%, with the possibility for further iterations and optimizations. This system can be implemented as an internal tool or as an externally outsourced product.

Results

The technology-based approach can drastically reduce incorrect product categorizations, which leads to a significant increase in findability and thus to an increase in sales. The ability to operate the system either internally or externally gives Rakuten flexibility in implementation and scaling.

Our Role

From the outset our team of experts was involved in the design, development and piloting of the technology-based approach. We worked closely with Rakuten to ensure that the prototype met the company specific requirements and delivered value in both the short and long term.