Mastercard Races to Outpace Fraud with AI-Powered ‘Decision Intelligence Pro’
Meta Description: Mastercard is leveraging sophisticated AI models, including its new Decision Intelligence Pro platform, to combat the escalating threat of fraud in a high-volume transaction environment.
The fight against financial fraud is increasingly a battle of scale, demanding ever-more-sophisticated defenses. Mastercard processes roughly 160 billion transactions annually, with peak periods – like the December holiday rush – seeing surges of 70,000 transactions per second. Identifying fraudulent activity within this torrent, without triggering false alarms, has proven a monumental challenge. Now, the company is turning to advanced artificial intelligence to gain an edge.
The Rise of Real-Time Fraud Detection
At the heart of Mastercard’s new strategy is Decision Intelligence Pro (DI Pro), a flagship fraud platform designed to analyze individual transactions in milliseconds. “DI Pro is specifically looking at each transaction and the risk associated with it,” a senior Mastercard official stated in a recent podcast. “The fundamental problem we’re trying to solve here is assessing risk in real time.”
DI Pro is engineered for speed and low latency. From the moment a consumer initiates a transaction – whether by tapping a card or clicking “buy” online – the data flows through Mastercard’s systems and onto the issuing bank in under 300 milliseconds. While the bank ultimately approves or declines the transaction, the quality of that decision hinges on Mastercard’s ability to deliver a precise, contextualized risk score.
Crucially, the system isn’t focused on identifying unusual activity, but rather on evaluating whether a transaction aligns with established consumer behavior. As one analyst noted, the goal is to understand if a purchase “makes sense” for the individual customer, asking, “Would we have recommended this merchant to them?”
An ‘Inverse Recommender’ Architecture
The core of DI Pro relies on a recurrent neural network (RNN), which Mastercard refers to as an “inverse recommender” architecture. This innovative approach frames fraud detection as a recommendation problem, with the RNN performing a pattern completion exercise to map relationships between merchants.
According to a company release, the system considers a user’s past behavior and current activity to determine the legitimacy of a transaction. This allows for a nuanced assessment beyond simple rule-based systems.
Mastercard is also addressing the complexities of data sovereignty – the principle that data is subject to the laws of the region where it’s collected – by utilizing aggregated, anonymized data that can be shared globally without raising privacy concerns. “So you still can have the global patterns influencing every local decision,” the Mastercard official explained. “We take a year’s worth of knowledge and squeeze it into a single transaction in 50 milliseconds to say yes or no, this is good or this is bad.”
Fighting Fire with Fire: Scamming the Scammers
The arms race between financial institutions and fraudsters is escalating, with criminals rapidly developing new techniques. Mastercard is responding by proactively engaging with cybercriminals on their own turf.
One tactic involves deploying “honeypots” – artificial environments designed to attract and trap malicious actors. When threat actors believe they’ve found a legitimate target, AI agents interact with them, aiming to uncover mule accounts used to funnel illicit funds. This approach is “extremely powerful,” according to the Mastercard official, as it allows defenders to use graph techniques to map connections between mule and legitimate accounts, ultimately revealing global fraud networks.
“It’s a wonderful thing when we take the fight to them, because they cause us enough pain as it is,” the official added.
Mastercard’s commitment to innovation extends beyond DI Pro, as highlighted in the podcast, which also discussed the creation of a malware sandbox with Recorded Future, the importance of a data science engineering requirements document (DSERD) for aligning engineering teams, and the need for relentless prioritization in AI deployment. Successful AI initiatives, the discussion revealed, require a three-phase approach – ideation, activation, and implementation – with activation often being overlooked.
