The startup Digital Materials, founded by Skoltech researchers (part of the VEB.RF group), together with the online hypermarket VseInstrumenty.ru, has developed an intelligent assortment segmentation solution that helps managers make purchasing and product matrix decisions based on demand, price, and product specifications. The main goal of the project was to solve a practical business challenge: developing a system for dynamic clustering of the product assortment according to demand and price segments based on sales data and product specifications.
The tool automatically identifies substitute products, reducing warehouse storage costs and procurement expenses without sacrificing assortment breadth. The solution is used daily by category managers and has been integrated into the purchasing planning process.
One of the tasks optimizing the management of the online retailer’s extensive assortment is clustering — the process of grouping products based on their specifications, prices, and consumer properties. For example, among thousands of power tool items, the system groups drills with similar power and voltage parameters, as well as similar prices, into one cluster so that managers can view them as interchangeable substitutes.
“Managers look at how different products are selling. Poorly selling items can be omitted from supplier orders — this simplifies logistics and storage. But it’s important to maintain a broad range: a low-selling product can be removed if there is a good substitute. Our approach allows us to automatically identify exactly such groups,” explained Dmitrii Maksimov, leader of the development project team.
Before the updated approach was implemented, assortment management considered only a limited set of technical specifications. This led to overly broad substitute groups — for example, low-power drills for home use and professional models could end up in the same cluster. This approach reduced recommendation accuracy and kept a high proportion of manual work when making decisions about delisting products. The Skoltech developers were tasked with creating a method that could automatically and efficiently process thousands of product categories and identify stable groups of interchangeable items within them.
“We applied an approach based on large language models. The algorithms automatically determine the most significant technical specifications for each product subcategory. We then use hierarchical clustering that combines categorical and numerical parameters, followed by grouping by price ranges. Our key focus was on automatically selecting meaningful features for each category. This approach allows us to consider not only the formal product specifications but also the economic logic behind their grouping,” shared Nikita Rybin, a senior research scientist at the Laboratory of Artificial Intelligence for Materials Design at the Skoltech AI Center and CEO of Digital Materials.
As a result, VseInstrumenty.ru was able to automate a process that previously required significant manual effort. The system automatically generates recommendations for delisting products and suggests alternative items. This has reduced assortment management costs, accelerated procurement decision-making, and eliminated risks associated with subjective parameter settings by managers.
“During our collaboration with the Digital Materials startup, we completely overhauled our warehouse product matrix. As a result, we accelerated assortment management decision-making several times over. This reduced storage costs while keeping the company’s turnover unaffected. All the goals we set for the project were achieved. I would particularly like to commend the professionalism of the development team — the results fully met our expectations,” commented Anastasia Naumova, the director of assortment and pricing management at VseInstrumenty.ru, the online hypermarket for professionals and businesses.
