Mastercard Unveils Gen AI Model as Insights Engine for Commerce

The company said its model, trained on billions of anonymized payment transactions, is built to generate insights across cybersecurity, loyalty programs, personalization, portfolio optimization and data analytics.

The move “reflects a broader shift we’ve been driving across Mastercard’s AI roadmap, moving beyond point solutions to foundation-level capabilities that learn from the complexity of global commerce,” Greg Ulrich, chief AI and data officer at Mastercard, said in an email to PYMNTS.

“In just the last month, we’ve advanced multiple efforts that show how AI can deliver smarter security, more relevant experiences, and stronger performance across the payments ecosystem.”

The initiative uses capabilities from Nvidia and Databricks.

From Language Models to Transaction Intelligence

Mastercard said that popular generative AI systems are based on large language models trained on unstructured data, while its approach relies on a large tabular model, a deep learning network using structured transaction datasets.

The company said it is training the model on multiple types of data, including payments transactions, merchant location data, fraud data, authorization data, chargeback data and loyalty program data. Mastercard said it plans to expand this work to include hundreds of billions of transactions.

The model works similarly to language models, which predict the next word in a sentence. In this case, the model predicts transaction characteristics based on historical data.

All training data is anonymized, with personal information removed before analysis.

“As the scale of digital transactions continues to explode past the billions each year, the financial services industry needs specialized AI models that can capture the full complexity and scale of global commerce in real-time,” Pahal Patangia, head of global industry business development for payments at Nvidia, added in the email to PYMNTS.

Cybersecurity as an Initial Use Case

Mastercard said cybersecurity is one of the first areas where it is applying the model.

The company said its existing cybersecurity AI models rely on data scientists to add features that help identify patterns such as spikes in purchase activity. The new model instead analyzes data with very limited human input and learns more independently which characteristics of the data are important.

In testing, the company said the model outperformed standard industry machine learning techniques. Mastercard said the model was better able to identify legitimate transactions, including high value but infrequent purchases such as buying a wedding ring, which can trigger false positives in existing systems.

The company said it plans to build hybrid cybersecurity systems that combine its existing AI models with the new foundation model.

The effort comes as payment networks face rising levels of AI-driven fraud. Mastercard has previously said that artificial intelligence is contributing to “unprecedented” levels of cybercrime while also becoming the primary defense against those threats, pointing to a future where AI systems increasingly counter one another across payment networks.

Expanding the Model’s Capabilities

Mastercard said the model could reduce the need to build, train, and maintain thousands of AI models for different markets, use cases, or customers. It is developing APIs and toolkits to give teams access to the model so they can build new applications.

The company added that it will continue expanding the model by incorporating additional datasets and enhancing its algorithmic sophistication. The model is being developed in line with Mastercard’s data responsibility principles, including user privacy, robust governance and controls and transparency.

Mastercard has also been expanding its AI efforts beyond this model. The company recently introduced Verifiable Intent, an open standard designed to verify transactions initiated by AI agents, aimed at helping issuers and merchants authenticate agent-driven payments within existing networks.

As reported by PYMNTS, the company also launched Agent Suite to help businesses integrate and manage AI agents across commerce workflows, including payments, fraud prevention and customer engagement.