When the text message came in, Kim Dow’s heart sank.
“Hi,” it read. “Did you just make this purchase with your REI Co-op Mastercard?”
The message went on to share the last four digits of Dow’s card number and a purchase for $88.69 worth of Alaska Airlines miles, which Dow says she did not make.
She texted back “no.” Within seconds, Mastercard’s fraud department disputed the charge and removed it from Dow’s active statement.
Though credit card safety technology has advanced in recent years with multi-factor authentication and EMV chips, fraud situations like Dow’s still happen. Mastercard, one of the country’s oldest credit card companies, is using artificial intelligence to prevent and minimize situations like this.
For the past 10 years, the credit giant has incorporated some form of machine learning algorithms to monitor transactions in real time and detect unusual patterns such as multiple failed logins and large or sudden withdrawals.
Mastercard’s newest iteration of its AI-powered fraud detection system features AI technology that scans nearly 160 billion transactions every year. This technology has helped Mastercard significantly reduce false-positive fraud cases, according to a May 2024 press release from the company.
AI powers a risk-scoring system that flags suspicious transactions
Identifying fraud often comes down to pattern recognition, which makes AI well-suited for the task, said Daryl Lim, an affiliate at the Center for Socially Responsible Artificial Intelligence.
“AI enables real-time detection of suspicious transactions by identifying patterns and anomalies impossible for human analysts to spot at scale,” Lim, who’s also the H. Laddie Montague Jr. Chair in Law at Penn State Dickinson Law, told BI.
This is especially true for a company such as Mastercard, which has large amounts of data that can be used to train an AI on what to look for, said Seckin Yilgoren, Mastercard’s senior vice president of security solutions in North American markets.
Yilgoren said the company processes nearly 160 billion transactions every year, providing it with terabytes upon terabytes of data to study and analyze for patterns that might expose fraud.
Mastercard uses this knowledge to inform a risk-scoring system called Decision Intelligence, which assigns a score to each transaction. To do this, the system continuously scans hundreds of millions of data points, such as a cardholder’s name, address, and purchase history, to predict whether a transaction is likely to be genuine.
Scores above a certain threshold are deemed legitimate, while those below the threshold are flagged as fraud. It does this in 50 milliseconds or less, Yilgoren said — an important point because it means the system can detect and block fraudulent transactions nearly in real time. This can prevent headaches for customers like Dow by giving them the opportunity to approve or decline suspicious transactions almost immediately.
Decision Intelligence is also continuously learning and adapting to new fraud patterns without human intervention, Yilgoren said. He added that this technology could snuff out false positives. By predicting context and behavior, AI can detect patterns the human brain cannot, and can distinguish between legitimate unusual activity, such as a rare entertainment splurge on a new television or concert tickets, and actual fraud, Yilgoren said.
This year, Mastercard also introduced Decision Intelligence Pro, a system that assesses the relationships between identifying user characteristics and past user behaviors to determine the validity of a transaction.
Lim told BI that these technologies make a compelling argument for AI as a method of fraud detection, but they also have limitations.
He said that machine learning models could inadvertently learn to associate legitimate transactions with fraud based on biased historical data. If this happens, an AI-powered system has the potential to flag certain demographics or locations disproportionately. This raises serious concerns around equity and trust, Lim said.
He said the final layer of fraud detection measures can often benefit from human judgment since they can investigate and provide insights into why any given transaction was flagged. “The goal is a hybrid model: AI for speed and scale, humans for nuance and accountability,” Lim told BI.
Andrew Reiskind, Mastercard’s chief data officer, agreed. He said the company has an AI governance program in which employees provide oversight on all AI-powered operations and solutions.
“By integrating human-centered design into our AI solutions and overseeing the process, we ensure that our technology not only enhances efficiency and delivers great products but also aligns with our ethical standards and commitment to responsible AI use,” Reiskind told BI.
Leveraging AI to fight fraud rings — and customer dishonesty
Mastercard also uses AI to detect fraud by mapping connections between accounts, devices, and transactions.
This technology uses behavioral biometrics, which examine how specific users type and swipe on apps, to try to detect imposters. Apps monitor this information as part of each transaction. From there, on the back end, an algorithm takes into account unique data points such as the cadence with which a user types in a password, how they hold a device, and how they move their mouse. Yilgoren said there is a “slight but noticeable” difference between a trusted user’s behaviors and the imitation of them.
There’s also the issue of first-party fraud, which occurs when a consumer makes a legitimate purchase and later requests a chargeback, even though goods and services were received. To combat this, Mastercard has rolled out its First-Party Trust program, which uses AI to scan data points such as IP address, device ID, and shipping address to determine the likelihood that a chargeback request is legitimate.
Yilgoren said this initiative incorporates AI and risk modeling to enable greater insight into a cardholder’s purchase history, delivery information, and geographic location. This, in turn, can help determine whether the original purchase was legitimate.
In 2024, Mastercard also launched Scam Protect, a suite of AI-powered solutions designed to help identify and prevent online scams.
“Really, it’s a question of how we can ensure data security and trust for our customers, but also for the banks and merchants who use our services,” Yilgoren said.