Razorpay Announces Early Results of Vulcan AI Payments Model
In a press release dated 13 November 2026, Razorpay, the Bengaluru‑based omnichannel payments platform founded in 2014 by IIT Roorkee alumni Shashank Kumar and Harshil Mathur, disclosed performance metrics from its transformer‑based foundation model, Vulcan. The model was trained on roughly three trillion data points derived from four billion payments and processes about three thousand signals for each transaction, leveraging infrastructure built with NVIDIA and AWS.
Merchant Impact
Businesses that have integrated Vulcan, including Blinkit, Bachatt and redBus, reported an eight to ten percent uplift in overall payment success rates. The AI‑driven routing and fraud engine also identified and stopped eight times more international card fraud and flagged five times more fraudulent or disputed transactions, all without generating additional alerts for merchants.
Shopper Experience
Razorpay noted that about forty percent more shoppers now see their preferred UPI application on the Razorpay Magic Checkout interface, a change that translates into an extra one to two lakh (100,000‑200,000) purchases each month across the platform.
Technical Advantages
Vulcan consolidates what were previously separate models for routing, fraud, risk and checkout, allowing it to draw on a broader set of transaction signals across merchants, payment methods, issuers and gateways. By evaluating roughly three thousand signals per payment, the model can select the healthiest routing path before a transaction is attempted and can flag cards used across unrelated sellers or risky cash‑on‑delivery orders before checkout.
Future Roadmap
Razorpay positions Vulcan as the initial layer of a continuously learning intelligence system that will eventually span authentication, routing, fraud detection and even lending decisions. The company cites India’s e‑commerce market, projected to reach US$350 billion by 2030, as a key driver for making digital payments more reliable, secure and intelligent.