Less is more – Daniel Schoess wins EHI/GS1 award
MSc MTEC graduate Daniel Schoess has won the EHI Foundation and GS1 Germany Science Award for his Master’s thesis in recognition of his innovative approach to predicting product interactions using minimal retail data.
Retailers constantly grapple with understanding how products influence each other on store shelves. Identifying whether items complement or substitute each other is essential for assortment planning, pricing strategies, and demand forecasting. However, predicting these interactions typically demands extensive, detailed data – something many retailers lack.
MSc MTEC graduate Daniel Schoess received the 2025 EHI Foundation and GS1 Germany Science Award for his Master's thesis, which he conducted at MIT with Professor Georgia Perakis and supervised by D-MTEC researchers Professor Florian von Wangenheim and Dr Sebastian Tillmanns. Schoess developed a prediction model for product interactions using minimal retail data that could significantly enhance assortment planning in retail.
The idea for the study emerged from a practical problem presented by his professor's industry partner at MIT. “They wanted to understand substitutional and complementary product effects using the simplest available data – basically, receipt data,” Schoess explains.
He leveraged his computer science background and connected the problem to existing recommender systems like those on Netflix and Amazon. These systems recommend products or services based on previous consumer behaviour, typically using comprehensive historical data. However, Schoess aimed to achieve similar predictive accuracy with only minimal transactional information.
“Retailers usually rely on customer-level data from loyalty programmes or online tracking,” Schoess explains. "But simple transactional data from cash registers is always available." He created so-called "product embeddings", virtual product representations derived from images and descriptions. The model learns to position these embeddings in a high-dimensional space based on sales receipt data to capture how products relate to each other.
“About sixty per cent of the products in our testing set were entirely new. Yet the model could predict their interactions, too.”Daniel Schoess
This addressed a significant limitation of traditional recommendation systems: the inability to pre-dict relationships for new products lacking historical sales data. “About sixty per cent of the prod-ucts in our testing set were entirely new,” Schoess says. “Yet the model could predict their interactions, too.”
His model identified complementary products – items frequently bought together – and distinguished them from substitutes effectively. “If someone buys a white T-shirt, they're unlikely to pick another identical T-shirt, but they might buy trousers that match,” Schoess explained.
“Retailers can integrate these predictions into assortment planning, pricing, and demand forecasting, even for completely new products,” Schoess says. “Thus, the method reduces retailers’ reliance on expensive customer data.”
Schoess experienced how challenging corporate data management can be himself. “I was used to working with tidy datasets,” he says. “However, working with large, raw corporate datasets required much more effort than anticipated, not only to clean the data but also to account for data sensitivity and infrastructure. The effort was necessary, though, to establish the method’s practical validity.”
After returning to Europe, Schoess started his doctorate at the ETH Zurich’s Center for AI Value, focusing on multimodal representation learning.
EHI Foundation and GS1 Germany Science Award
The EHI Foundation and GS1 Germany Science Award honours researchers and startups for their scientific work and projects relevant to the retail and consumer goods industries. The prize, endowed with a total of 65’000 euros, is awarded in the categories start-up, doctoral thesis, and Master’s thesis.
Always up to date
Would you like to always receive the most important internal information and news from ETH Zurich? Then subscribe to the "internal news" newsletter and visit Staffnet, the information portal for ETH employees.