Emerging evidence
More
8x More International Card Fraud Stopped: Early Results from Razorpay's Vulcan AI Payments Model
- Evidence type
- News report
- Sources
- 1 sourceJournalism
- Published
- 2 min read
- Truvace record
- TRV-2026-2897cite or check it
Businesses including Blinkit, Bachatt and redBus report higher payment success and sharper fraud detection from Razorpay's payments foundation model Bengaluru, Karnataka, India (NewsVoir) Since announcing Vulcan, its transformer-based foundation model for payments, Razorpay is sharing what the model has delivered...
The gain
Razorpay reported that merchants including Blinkit, Bachatt and redBus using components of its Vulcan payments model saw 8-10% higher payment success rates and 8x more international card fraud detected and stopped.
In brief
Razorpay, an Indian payments company, says components of Vulcan, its transformer-based payments foundation model, are already running on its network and informing routing, fraud and risk decisions in real time. Merchants including Blinkit, Bachatt and redBus reported 8-10% higher payment success rates and 8x more international card fraud detected and stopped.
Vulcan was trained on about 3 trillion data points across 4 billion payments and built with NVIDIA and AWS. It uses roughly 3,000 signals per transaction across merchants, payment methods, issuers and gateways, replacing separate models for routing, fraud, risk and checkout that do not share signals.
Razorpay says failed payments matter most to first-time and small-town shoppers, for whom one failure can mean going back to cash. The company plans to extend the model to authentication and lending. The results come from merchants and the company itself, and no independent evaluation is described.
Main points
- Merchants using components of Vulcan reported an 8-10% improvement in payment success rates.
- They also reported 8x more international card fraud detected and stopped.
- Merchants identified 5x more fraudulent or disputed transactions without an increase in the number of alerts.
- 40% more shoppers see their preferred UPI app on Razorpay Magic Checkout, helping complete 1-2 lakh more purchases every month.
- Vulcan was trained on about 3 trillion data points across 4 billion payments and built with NVIDIA and AWS.
- The model draws on roughly 3,000 signals per transaction and can flag a card used across unrelated sellers or warn merchants about risky Cash on Delivery orders before checkout.
The full story
Razorpay says Vulcan combines tasks that the industry has traditionally handled with separate specialised models: routing, fraud, risk and checkout. Because those models do not share signals, patterns visible across merchants, issuers and gateways can be missed when payments are judged one at a time.
In India a single purchase can run through UPI, cards, net banking, wallets or Cash on Delivery across hundreds of banks and gateways. Vulcan aims to route each payment down the healthiest path before it is attempted, so a card, bank account and funds that are all fine do not produce a failed payment.
Razorpay cites an internal study of 1.5 million shoppers and 51,000+ businesses that found failed transactions, drop-offs and delays in metros and small towns alike. It plans to extend Vulcan to authentication, routing, fraud detection and lending, and cites a projection of India's e-commerce market reaching $350 billion by 2030.
Sources
1 on file- The Tribunetribuneindia.com
Talk it over
Merchants including Blinkit, Bachatt and redBus reported 8x more international card fraud stopped. What would you want to know about how those figures were measured before trusting them?
8x More International Card Fraud Stopped: Early Results from Razorpay's Vulcan AI Payments Modelhttps://truvace.com/item/8x-more-international-card-fraud-stopped-early-results-from-razorpays-vulcan-ai-
Merchants including Blinkit, Bachatt and redBus reported 8x more international card fraud stopped. What would you want to know about how those figures were measured before trusting them?
Answer in the threadRazorpay says failed payments matter most to first-time and small-town shoppers, for whom one failure can mean going back to cash. Who benefits most from Vulcan, and who might see little change?
Answer in the threadVulcan uses roughly 3,000 signals per transaction to flag risky orders. How would you weigh the higher payment success rates against shoppers having so much of their payment data read by one model?
Answer in the thread
If you have more time
The results come from Razorpay and the merchants using Vulcan components. Which independent party should test them, and what baseline and time period would you ask them to use?
Answer in the threadRazorpay plans to extend Vulcan to authentication and lending. How would decisions about loans differ from decisions about routing a payment, and who should oversee them?
Answer in the thread
ace




What readers say