Scientists from Sber, AIRI, ISP RAS, and the Steklov Institute of Mathematics of the Russian Academy of Sciences have developed an approach that helps AI see not just a single user action, but the entire picture of their digital connections. This can lead to more accurate fraud detection, product and banking service recommendations, and risk assessment.

Traditional models often analyze a client individually: what they bought, where they clicked, how long ago they accessed the service. The new technology adds connections with other people, products, and categories to this. This allows the algorithm to detect suspicious patterns that are not visible when analyzing a single account.

The method was tested on four large datasets from the financial sector and e-commerce. The maximum increase in accuracy by the AUC metric was 2.3%.

The development will be useful for banks in combating fraud, for marketplaces in providing more accurate recommendations, and for digital platforms in predicting customer churn. The research was accepted at The ACM Web Conference 2026 – one of the major international conferences on web technologies and AI.