Fintech unicorn Razorpay has launched Vulcan, an AI foundation model developed specifically for the payments ecosystem, with technology support from NVIDIA and AWS.
The Bengaluru-based company said Vulcan has been trained on nearly 3 trillion data points across 4 billion payments. The transformer-based model uses around 3,000 signals per transaction to make payment-related decisions and is designed to improve how businesses manage payment routing, fraud detection, risk assessment and checkout experiences.
AI Model Designed for Payment Decisions
Razorpay said Vulcan acts as a common intelligence layer across multiple payment functions. Instead of relying on separate models for different use cases, the system is designed to bring these capabilities together through a single AI foundation model.
The model can analyse signals from merchants, payment instruments, issuers and gateways to assess different payment routes and select the one most likely to succeed.
It can also identify fraud patterns across merchants, assess risks linked to cash-on-delivery orders and recommend suitable payment methods to shoppers during checkout.
NVIDIA and AWS Support Vulcan’s Development
NVIDIA GPUs were used to train and run Vulcan, while AWS cloud infrastructure and Amazon SageMaker supported its development, training and deployment.
Razorpay said the model has been built from the ground up, with its architecture and training data remaining proprietary to the company.
Several components of Vulcan were already being tested across Razorpay’s payment network before the official launch. Customers including Blinkit, Bachatt and redBus have started using some of its capabilities in live payment environments.
Razorpay Reports Higher Payment Success and Fraud Detection
According to Razorpay, Vulcan has helped improve payment success rates by 8-10% and detect and stop 8x more international card fraud.
The company also said the technology has identified five times more fraudulent or disputed transactions without increasing the number of alerts generated.
Through Magic Checkout, Razorpay said 40% more shoppers are now shown their preferred UPI app. This has helped complete an additional 1-2 lakh purchases every month, according to the company.
Built From Billions of Payment Signals
Vulcan uses a broad set of signals generated across the payment ecosystem to make decisions in real time. These include information from merchants, payment instruments, issuing banks and payment gateways.
The model can use these signals to determine which payment route is most likely to succeed, while also identifying patterns that could indicate fraudulent activity.
Beyond fraud and routing, the system can evaluate risks associated with cash-on-delivery transactions and recommend payment methods based on shopper behaviour during checkout.
Razorpay’s internal study involving 1.5 million shoppers and more than 51,000 businesses found payment friction across both metropolitan and smaller markets. The company identified issues including failed transactions, delays and customer drop-offs.
Vulcan to Expand Into Lending and Authentication
Razorpay plans to extend Vulcan’s capabilities to additional areas, including authentication and lending, as it works towards using a single AI layer for a wider range of payment-related decisions.
The launch comes as India’s digital ecommerce market continues to expand. Razorpay estimates that the country’s digital ecommerce market could reach $350 billion by 2030.
With Vulcan, the fintech unicorn is looking to use its large payment data network to build AI capabilities that can improve transaction success, reduce fraud and make digital payments more personalised for both merchants and consumers.
-By Shivani Solanki
