Graph neural networks help identify suspicious transfers through connections between clients

Associate Professor and Head of the Skoltech-Sberbank Applied Research Laboratory Aleksey Zaytsev, in an interview with TV BRICS, commented on data concerning the implementation of artificial intelligence (AI) in South Africa’s financial sector. It was previously reported that more than half of the country’s major banks invested over 20 million South African rand (around US$1.26 million) in AI systems to combat fraud in 2024.

According to the expert, such investments cannot be described as a unique breakthrough – this is a natural process taking place in all healthy markets. He identified graph neural networks as the most promising area in combating fraud. These analyse not only the actions of a specific client but also that client’s connections with other individuals.

“It is the client’s interaction with other people that matters: to whom money is transferred, in what amounts, and how typical these behavioural patterns are. […] The set of connections between clients forms a graph of relationships around the client. […] Analysing these connections helps to detect fraud even in cases where the system has not previously encountered this type of deception,” Zaytsev explained.

As an example of a country where such technologies are developing most rapidly, the expert cited China. There, a large number of banks and artificial intelligence researchers ensures the swift advancement of technology. In Russia, he noted, trends are similar. A strong community of machine learning specialists has emerged in the banking sector, and they have long been developing comparable models.

Responding to a question about the exchange of technologies and best practices among BRICS countries, Zaytsev pointed out that professional communication mainly takes place at the local level – at industry conferences within a single country. More global discussions of methods occur at academic venues.

“If we speak about something more global, there are now quite a number of international conferences on machine learning. Among the leading ones related to BRICS, for example, one can mention the Asian Conference on Machine Learning. It has a very strong line-up of participants, professors […] If we are talking about academic conferences, they discuss the latest methods and provide an understanding of where everything is heading and which methods work and which do not,” Zaytsev said.

The expert also commented on the use of artificial intelligence for the personalisation of financial services. According to him, the technologies make it possible not only to identify fraudsters but also to better understand clients’ needs. At the same time, Zaytsev acknowledged that the technologies remain imperfect.

The expert noted that banks in different countries address the task of combating fraud in broadly similar ways. While external tools are often used to assess a borrower’s creditworthiness, banks rely on their internal solutions to protect against fraudsters.

According to a report by the Financial Sector Conduct Authority (FSCA), artificial intelligence in South Africa’s financial sector is used not only for security but also for the personalisation of services. Algorithms help to offer clients tailored loans and insurance policies, while dynamic pricing is being introduced in the insurance market – premiums may change in real time based on data from vehicle telematics or fitness trackers. At the same time, regulators also highlight risks: threats to data privacy, potential algorithmic bias and systemic vulnerabilities. In this regard, the FSCA plans to develop approaches to regulating AI technologies in the financial sector.

African Times published this article in partnership with International Media Network TV BRICS