What It’s Like to Build Generative AI Models at Mastercard

 There is a golden rule when it comes to AI: good data in, good data out. 

That’s especially true when building a model on billions of anonymized transactions, which is what the AI team at Mastercard is doing.  

“We plan to use this new foundation model — not to build a chatbot — but as an insights engine that will make many of our tools and services even better, from cyber defenses to loyalty programs to small business tools,” Distinguished Engineer Steve Flinter said in a recent Mastercard blog post.

The new generative AI model is based on a large tabular model (LTM), a type of deep learning neural network that uses structured data like datasets or large-scale tables at a scale comparable to large language models (LLM). 

Flinter said that Mastercard’s plan is to “ramp up this work” to include hundreds of billions of payment transactions, as well as additional types of datasets like merchant location data, fraud data, authorization data, chargeback data and loyalty program data.

 

 

The new generative AI model marks a significant milestone for the global technology company’s impact on commerce, with potential applications spanning cybersecurity, fraud detection, personalization, and small business tools. 

To make this work possible, Flinter said Mastercard is leveraging Nvidia, an accelerated computing and AI infrastructure company, and Databricks, a data and AI platform company.

“We’re already seeing strong results,” Flinter said, citing better fraud detection capabilities than previous models. 

Mastercard showcased its work during the Nvidia GTC 2026 conference, where business leaders, developers, researchers and students gathered to learn about real-world use cases of AI.

For engineers and AI professionals looking for their next career challenge, Mastercard offers the unique chance to build and optimize scalable, AI-driven solutions that could transform the safety and efficiency of payments for consumers and major brands all over the world.

 

Why Mastercard Built a Large Tabular Model Instead of an LLM

Most people are familiar with LLMs, which are built using large amounts of unstructured data like text and images. Think ChatGPT, Claude and Gemini — models built for language-heavy tasks.

Mastercard’s new generative AI tool is different, and for good reason. 

As Flinter explained in a Mastercard blog post, an LTM made more sense for the payment technology company because payments data is mostly structured, tabular and event-based; for example, transaction records, fraud incidents, merchant and location data.

Powered by Nvidia’s accelerated computing platform, Flinter said this data is being processed at “unprecedented” speeds.

“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,” Flinter said in the blog post.

Mastercard’s new LTM doesn’t mean the company is forsaking its current AI models, however. 

In fact, part of Mastercard’s AI roadmap is building hybrid cybersecurity systems that combine their current models with the new foundation model’s capabilities to “future-proof” their cyber defenses, Flinter added. 

“Our new foundation model analyzes the same data with very limited human input as a starting point, learning more independently what the important characteristics of the data are,” Flinter said in the blog post. “In this way, the LTM could identify new connections in the data that a human might not find on their own.”

 

How Mastercard Uses Structured Transaction Data to Train Its AI Model

Mastercard’s newest AI model stands out not only in how it was built, but also in the capabilities it’s expected to deliver.

“In our testing, we’ve already seen this new model outperform standard industry machine learning techniques, giving us promising early signs,” Flinter said in the blog post. 

“In our testing, we’ve already seen this new model outperform standard industry machine learning techniques, giving us promising early signs.”

The engineer gave the example of a false positive cybersecurity alert when a big but infrequent purchase is made, like a wedding ring. With the new model, Mastercard can better identify valid transactions from weaker data signals. The new model will also allow Mastercard to maintain fewer AI models, as the new LTM is more flexible.

“This cybersecurity example is just one potential outcome of this research,” Flinter said in the blog post. “We believe the new foundation model can also be used to improve loyalty and rewards programs, personalization models, portfolio optimization and data analytics tools.”

 

How Mastercard Plans to Expand Its AI Model Across the Business

While the new LTM’s announcement is a major milestone for Mastercard, Flinter said in the blog post that it’s just the beginning. The Mastercard team is already looking to expand the LTM’s capabilities, adding algorithmic sophistication to the model’s architecture so it can glean even more insights from raw data.

“Plus, we’re developing APIs and toolkits to give teams across Mastercard access to this new foundation model, so they can build new applications on top of it,” Flinter said in the blog post.

In the future, the LTM could be used to enhance rewards programs by refining how the company identifies relevant offers for cardholders. It could enhance personalization models, leading to more accurate predictions of customer behavior.

Greg Ulrich, Mastercard’s chief AI and data officer, told PYMNTS that Mastercard’s path toward AI advancement reflects a broader shift it has been driving — beyond point solutions to “foundation-level capabilities that learn from the complexity of global commerce.” 

At Mastercard, AI is now a core part of how the company builds products and services. For engineers, data scientists and machine learning professionals who want to work on AI that matters in production — not just demos — Mastercard is the place to be. 

 

  

Frequently Asked Questions