Tinkoff has developed a model based on machine learning, which allows fraud to be detected at the stage of confirming a loan application, said Oleg Zamiralov, deputy head of the Tinkoff Center for Ecosystem Security, at the Ural Cybersecurity Forum in finance With the help of technology, it is possible to stop over 90% of cash loan applications processed under the influence of social engineering.
Scoring, which was previously used to check a customer's creditworthiness, has been enhanced with features and parameters that indicate potential fraud. To do this, several million applications were analyzed, including applications that clients submit under the influence of social engineering. For example, such signs include socio-demographic factors, anomalies in the application itself, characteristic of scammers’ instructions, and much more. This model is capable of accurately predicting the impact of third parties when a client fills out an application and sends a signal to bank employees so that they double-check the application and, if fraud is suspected, reject it.
This creates an additional layer of protection to existing ones, such as interrupting a fraudulent call, blocking suspicious transactions when the fraudster is already trying to withdraw credit funds from the account, and others. The peculiarity of this layer of protection is that the bank does not need to convince the client to make a transaction to the fraudster.
“We periodically encounter situations when the bank understands that the client is under the influence of social engineering, but the client himself does not realize this, does not believe the bank employees who are trying to convince him, and insists on making a transfer. The technology helps prevent fraud even at the stage of the loan application: if we understand that the client is taking out a loan against his own will, we simply reject this application.
We have been testing this pilot since the end of last year, and it shows high efficiency — in particular, with the help of this model we manage to stop over 90% of cash loan applications processed under the influence of fraudsters,” notes Oleg Zamiralov.
Another preventive method of combating fraudulent loans based on artificial intelligence is the development of Victim Score. This is a special rating that determines, based on various signs, that the client is at risk — with a high probability of believing scammers. The system automatically begins offering additional training to such clients in stories and in the blog on the security of the bank’s mobile application. Victim Score works especially effectively using a pseudo-investment scheme. “According to our research, we see that those who have undergone training are 2–2.5 times less likely to believe scammers than those who have not undergone training. In particular, among the entire group of trained clients, the percentage of fraud is 0.8%, and among untrained clients it is already 1.5%,” adds Oleg Zamiralov.
Свежие комментарии