Key Takeaways
- Razorpay launches Razorpay Vulcan, an AI foundation model built specifically for payments.
- Vulcan analyzes 4 billion payments and about 3 trillion data points.
- The model targets payment failures, fraud detection, risk assessment, and checkout personalization.
Razorpay launched Vulcan on Tuesday, an AI foundation model built for payments, to improve transaction success, detect fraud and personalize checkout experiences across India’s digital payments network.
Vulcan Trains On 4 Billion Payments
Razorpay said Razorpay Vulcan is India’s first transformer-based AI foundation model designed specifically for payments. Built using technology from NVIDIA and Amazon Web Services, the model has been trained on approximately 3 trillion data points across 4 billion payments.
The model analyzes about 3,000 signals from each transaction. Razorpay said early components are already operating across its payments network, with companies including Blinkit, Bachatt and redBus using the technology in live payment environments.
Razorpay CEO and co-founder Harshil Mathur said the company developed Vulcan to address reliability problems in India’s fast-growing digital payments market.
In an interview with Hindustan Times, Mathur said early testing involving more than 1.5 million transactions across more than 50,000 merchants showed an “8 to 10% improvement in payment success rates.”
Model Targets Fraud And Payment Failures
Razorpay said early applications of Razorpay Vulcan have improved payment success rates by 8% to 10% while detecting eight times more international card fraud. The company also said the system identified five times more fraudulent or disputed transactions without increasing the number of alerts.
The model is designed to handle several payment functions through a shared system instead of relying on separate machine-learning models. These functions include payment routing, fraud detection, risk assessment, and checkout personalization.
For payment routing, Vulcan can assess a transaction in real time and select the path it predicts is most likely to succeed. It can also identify potentially risky cash-on-delivery orders and recommend payment methods based on customer behavior.
Razorpay said Razorpay Vulcan has improved its Magic Checkout experience, with 40% more shoppers seeing their preferred UPI application. The company estimates that this has contributed to an additional 1 lakh to 2 lakh purchases each month.
Razorpay Plans Broader AI Use
The company said Vulcan is proprietary and was built from the ground up, with Razorpay owning its architecture and training data. NVIDIA GPUs supported model training and deployment, while AWS infrastructure, including Amazon SageMaker, supported development and deployment.
Razorpay said the model is intended to eventually support decisions involving authentication, payment routing, fraud detection and lending. The company is developing the system as India’s digital commerce market expands.
Vulcan is being introduced as India’s payment ecosystem continues to grow across UPI, cards, net banking, wallets and other methods. Razorpay said its own research involving 1.5 million shoppers and more than 51,000 businesses identified recurring problems involving failed payments, delays and customer drop-offs.
The company said Razorpay Vulcan is designed to learn patterns in payment activity rather than text, distinguishing it from general-purpose language models.
Visit CyberPro Magazine For The Most Recent Information.
