What Mastercard Users Can Expect of the Firm’s New AI Engine

Mastercard has researched and built a large-scale AI model, powered by NVIDIA and Databricks technology, that can be used as the basis for a wide range of applications.

The credit card company is using the NVIDIA NeMo AutoModel and NVIDIA accelerated computing together with the Databricks platform to develop its proprietary transaction foundation model, one of the first payments-specific models to understand the nuances of global commerce.

The model is trained on hundreds of millions of transactions and is already showing promising signs of outperforming advanced machine learning techniques across operations.

“Our new foundation model is a different kind of deep learning neural network, called a large tabular model, or LTM, which is trained on structured data, such as large-scale tables or datasets,” says Steve Flinter, Distinguished Engineer at Mastercard.

Mastercard plans to train the model further on more payments transactions, as well as additional types of datasets including merchant location, fraud, authorisation and chargeback.

“As we train the model on more data and more kinds of data, it will be able to provide more insights and predict future transactions with greater accuracy,” says Steve.

First thing’s first: improving security

Mastercard’s priority is to use the model to improve its cybersecurity.

The large-scale AI model is built with NVIDIA and Databricks technology, and is a type of LTM that learns key characteristics and patterns with very limited human input.

Unlike existing security models, which require data scientists to manually add features to raw transaction data to identify things like a sudden spike in purchasing, this new LTM analyses data independently, identifying new connections that a human might not find.

This improved understanding has already shown promise in reducing errors, specifically by better identifying legitimate, very expensive but infrequent purchases, such as a wedding ring, which current models often incorrectly flag as fraud (a “false positive”).

Beyond security

Cybersecurity is just one potential outcome of using the LTM. It can also be used to improve rewards programmes, personalisation models, portfolio optimisation and data analytics tools.

It could also become flexible enough to help the firm cut down the thousands of AI models it maintains for different markets, use cases and customers.

Industry success

Other finance firms are seeing success using similar technology. For instance, Revolut built a transaction foundation model using a self-learning method called masked prediction to improve fraud detection and accurately predict a customer’s next purchase.

The company achieved these results by using NVIDIA’s full AI stack – including NVIDIA Hopper GPUs, the NVIDIA cuDF library and the NVIDIA Nemotron family of open models – resulting in a 20% increase in fraud detection precision, better credit risk predictions and a 9.6% uptick in cross-sell accuracy.

What’s next

Mastercard is planning to enhance the internal programming of its new foundation model. By improving the model’s structure, it aims to allow it to find deeper, more complex patterns and relationships within the payments data it analyses

Plus, it is working on training teams across Mastercard to access this new foundation model, so they can build new applications on top of it.

Company portals

Executives