Mastercard’s AI Garage is building products for the world in India: Nitendra Rajput

Mastercard’s AI Garage is building global payment network products for the world in India, Nitendra Rajput, Senior Vice-President and Head, AI Garage, India at Mastercard, told The Economic Times Digital during the Global Fintech Fest (GFF) 2026.

Launched in 2018, the AI research and development hub applies machine learning and data science to secure global payment networks. It is primarily based in Gurugram.

“Since the business, from a revenue perspective, is significantly high outside of India, the majority of our products are used by customers outside of India. But some of them are for India as well,” Rajput said.


Also Read: Making people redundant is a no-win: Revolut India CEO on AI job cuts

The evolution of AI garage happened in phases: post the initial investment, a lot of trials and errors were conducted to test for the integration of AI (HR, finance, operations, etc) and identify opportunities of efficiency gains.

He added, “When it started to expand, we started to question… how can we actually generate revenue out of this? Building things that are efficient is one thing, but building things that are so useful that someone is going to pay for it is a different game. That's when we started to get into the real Mastercard business, because that is all about transactions.”

Rajput claimed that Mastercard does about 160 billion transactions a year and AI is treated as a tool to verify the authenticity of merchants and then take actions on the transactions.

Explaining the business model, Rajput said, “Banks lose a lot of money with fraudulent transactions. In order to reduce that loss, they can pay a company like Mastercard to give them insights and save on the fraud. That's the revenue-generating model for us.”

Mastercard has a product called Merchant Risk Predict (MRP), which predicts the risk of a merchant being bad or fraudulent and prevents the transaction from going through when a user attempts to pay through their Mastercard for online purchases.

Large Tabular Model over LLM

Earlier this year, Mastercard announced that it has tied up with NVIDIA and Databricks to build a new foundation model or a deep learning neural network, called a large tabular model or LTM. The LTM is trained on structured data, such as large-scale tables or datasets, and underlying concepts around graph neural networks and sequential modeling, among others.

Explaining LTM's advantage over an LLM, Rajput said, “That's (LTM) is more relevant to us because we don't have textual data; we have tabular data. So, you build your own model, which is trained on transaction data, and not use any other closed-source models.”

Also Read: BFSI sector in no hurry to adopt complete automation: Perfios Group CEO

He claimed the unavailability of a vertically-deep model and the unique characteristics of Mastercard’s data were the primary factors behind building an in-house LTM. “Our data has very unique characteristics…. such as an amount, a time, a type, whether you had tapped or swiped or did an online transaction, and so on. Any existing external model doesn't have these features anywhere, so you would have to work on building a model on these specific features for us.”

In addition to an LTM, Mastercard forayed into agentic commerce via payment aggregators such as Cashfree Payments, Juspay, PayU and Razorpay in March this year. But, it is yet to go live for users in India.

“I think the regulations in India, whenever that happens, when they start allowing agentic transactions, that's when they would,” noted Rajput, without divulging further details.

Establishing guardrails

Rajput said Mastercard’s AI governance framework covers how and where data is used, as well as whether models continue to perform fairly after deployment. The company checks models for fairness during development and monitors them once they go live to identify any deviations, he said.

“We have to ensure that the data being used is the one that is allowed and it complies with the regulations,” Rajput said.

On accountability, Rajput said the industry will need to establish who bears liability when an AI agent makes a decision that leads to a disputed transaction, whether that is the issuer, merchant or acquiring bank. He added that, from a technology perspective, financial institutions must be able to understand why an AI agent arrived at a particular decision, especially in a regulated environment.

Rajput said Mastercard had demonstrated “reasoning as a service” at the India AI Summit, where an LLM was designed to provide an explanation for the decisions it made.

Irrespective of stringent regulations, India is a growth hub for Mastercard as the fintech ecosystem continues to evolve aggressively. “The amount of innovation in general that you see, that is happening in the fintech world in India is significantly fast, as opposed to any of the geographies that you would see. So the whole ecosystem is very supportive, very receptive to wanting to try stuff,” he noted.