Mastercard's DI Pro: How AI Fraud Models Process 160 Billion Transactions Annually in Under 300 Milliseconds
Mastercard's Decision Intelligence Pro (DI Pro) platform leverages sophisticated AI models to detect fraudulent transactions in real-time, processing roughly 160 billion transactions a year and handling surges of 70,000 transactions per second. Designed for speed and low latency, DI Pro assesses the risk of individual transactions within 300 milliseconds, providing issuing banks with precise, contextualized risk scores. Unlike traditional anomaly detection, DI Pro's core is an "inverse recommender" recurrent neural network (RNN) architecture that treats fraud detection as a recommendation problem, identifying patterns in merchant relationships to pinpoint suspicious activity. This system aims to solve the fundamental problem of real-time risk assessment in a race against the scale of fraud.
Fraud protection is a constant battle against immense scale. Mastercard's network, for example, handles approximately 160 billion transactions annually, experiencing peak surges of up to 70,000 transactions per second, particularly during busy periods like the December holidays. Identifying fraudulent purchases amidst this volume, without generating excessive false alarms, is a formidable challenge that fraudsters have historically exploited. However, advanced AI models are now capable of scrutinizing individual transactions to pinpoint suspicious activity within milliseconds. This capability is central to Mastercard's flagship fraud platform, Decision Intelligence Pro (DI Pro).
Johan Gerber, Mastercard’s EVP of security solutions, explained in a recent VB Beyond the Pilot podcast that "DI Pro is specifically looking at each transaction and the risk associated with it." He emphasized that "The fundamental problem we're trying to solve here is assessing in real time."
DI Pro was engineered with latency and speed as primary considerations. From the moment a consumer initiates a transaction, whether by tapping a card or clicking 'buy,' the data flows through Mastercard’s orchestration layer, back onto the network, and then to the issuing bank. This entire process typically completes in less than 300 milliseconds. While the ultimate decision to approve or decline rests with the bank, the quality of that decision is heavily reliant on Mastercard's ability to deliver an accurate, contextualized risk score indicating potential fraud.
A key complexity in this process is that the system isn't merely searching for anomalies. Instead, it's designed to identify transactions that, by their very nature, mimic legitimate consumer behavior. At the heart of DI Pro is a recurrent neural network (RNN) that Mastercard refers to as an "inverse recommender" architecture. This innovative approach reframes fraud detection as a recommendation problem. The RNN performs a pattern completion exercise to understand how merchants relate to one another. As Gerber elaborated, the system considers factors such as: "Here's where they've been before, here's where they are right now."