Mastercard implemented AI to detect fraudsters
Payment system Mastercard has announced the creation of a generative artificial intelligence model to help banks detect fraudulent transactions in their networks. The Decision Intelligence Pro II model is designed to evaluate suspicious transactions in real time and determine their legitimacy.
Ajay Bhalla, president of Mastercard’s enterprise risk and security division, said the new solution is a proprietary recurrent neural network built from scratch by the company’s cybersecurity and anti-fraud teams.
Mastercard uses transformative models to help harness the full power of generative AI, Bhalla notes. According to him, all data used in the model is taken from the company’s ecosystem.
In some cases, Mastercard relies on open source when necessary, but much of the technology is built in-house, Bhalla explained.
The Decision Intelligence Pro algorithm is trained on the data of approximately 125 billion transactions that pass annually through the company’s card network. This information helps the neural network understand relationships between clients, rather than words, as big language models like GPT-4 and Gemini from OpenAI and Google do.
Instead of text input, Mastercard’s algorithm uses the cardholder’s history of store visits. In this way, the model matches whether the transaction matches the customer’s previous transactions by location. The algorithm then generates paths through the Mastercard network to find the answer as scores. A higher score will correspond to a pattern of normal behavior expected of a cardholder, while a lower score reflects being outside that pattern.
According to Mastercard, this entire process takes 50 milliseconds. Bhalla believes that the technology will help financial organizations improve the detection of fraudulent transactions by an average of 20%. In some cases, the model led to a 300% increase in fraud detection, the expert summarizes.
Over the past five years, Mastercard has invested more than $7 billion in cybersecurity and AI technologies. Last year, the payment system acquired the Swedish cybersecurity company Baffin Bay Networks.
Mastercard believes that Decision Intelligence Pro will allow banks to save up to 20% of expenses by eliminating most of the costs that are usually spent on evaluating illegal transactions. The potential of the model lies in its ability to detect fraudulent patterns and trends to predict future types of crime that are currently unknown in the payment ecosystem.