AI Reduces Credit Card Fraud by 300%
Mastercard has announced the development of a cutting-edge generative artificial intelligence (AI) model, designed in-house to enhance fraud detection across its extensive network of banking partners. This innovative AI tool, named Decision Intelligence Pro, aims to provide banks with the capability to more accurately evaluate suspicious activities on Mastercard’s network, facilitating real-time determinations of transaction legitimacy.
In an exclusive dialogue with CNBC, Ajay Bhalla, the president of Mastercard’s cyber and intelligence business unit, shared insights into the technology underpinning this advancement. According to Bhalla, Decision Intelligence Pro leverages a bespoke recurrent neural network, a key element of generative AI technology, meticulously crafted by Mastercard’s cybersecurity and fraud prevention teams. This model incorporates transformer models to harness generative AI’s capabilities, offering a robust solution constructed entirely in-house utilizing diverse data sources within the company’s ecosystem.
“We are using the transformer models which basically help get the power of generative AI,” Bhalla told CNBC “It’s all built in house we’ve got all kinds of data from the ecosystem. Because of the very nature of the business we are in, we see all the transaction data which comes to us from the ecosystem.”
Bhalla revealed the strategic use of open-source resources as necessary, but the bulk of the technology was developed internally. The proprietary algorithm undergoes training with data from approximately 125 billion transactions processed through Mastercard’s network each year. This vast dataset enables the AI to discern patterns between merchants, focusing on transactional relationships rather than textual analysis, to pinpoint potential fraud.
Mastercard’s algorithm uniquely assesses the likelihood of a cardholder visiting a merchant involved in a transaction, using merchant visitation history as a basis for analysis. This approach generates a predictive score through the network, akin to a heat-sensing radar, to gauge transaction authenticity within an impressive timeframe of just 50 milliseconds.
Bhalla highlighted the technology’s significant impact, with financial institutions witnessing an average fraud detection rate improvement of 20%, and in some cases, up to 300%. Over the past five years, Mastercard has allocated over $7 billion towards cybersecurity and AI technologies, including strategic acquisitions like the purchase of Swedish cybersecurity company Baffin Bay Networks in March 2023.
The company’s efforts in AI innovation parallel those of its competitors, such as Visa, which recently established a $100 million venture fund focused on generative AI startups in October 2023. Mastercard anticipates that its AI model will not only enhance fraud detection capabilities but also offer substantial cost savings for banks by reducing expenditures related to the investigation of fraudulent transactions.
Bhalla envisions the technology’s long-term potential to identify and predict emerging fraud trends within the global payment ecosystem, leveraging the comprehensive transaction data from Mastercard’s worldwide customer base. This development comes amidst a wave of AI-driven innovations in the payments and digital banking sectors, highlighted by companies like PayPal introducing new AI-based features and enhanced checkout processes.
Certain types of AI get the majority of attention, namely the Large Language Models (LLM) like ChatGPT and Grok, as well as AI art. But there is actually a whole world of AI applications and innovations that do (generally) more specific and useful tasks. We have covered some of these awesome applications such as advanced farming that could completely change monoculture farming, to a AI fire-detection system that reduced forest-fires by an amazing amount.
Overall, this is a fantastic example of AI being used for something good! Having experienced credit card fraud myself last year, I’m really glad to see Mastercard addressing this issue.