.png){kind=logo}
Vulcan: Razorpay trained Vulcan on 3 trillion data points from 4 billion payments.
Payment success: Early Vulcan deployments show an 8–10% improvement in payment success.
Fraud detection: Razorpay reports 8× more international-card fraud detected and stopped.
Razorpay now uses AI across several parts of the payment journey, from checkout and payment routing to fraud checks and authentication. Its latest move is Vulcan, a transformer-based AI foundation model built for payments.
Razorpay trained Vulcan on about 3 trillion data points from 4 billion payments. The model can study around 3,000 signals for each transaction. These signals can include payment behaviour, customer patterns, merchant activity, device details, payment routes and risk indicators.
The scale gives Razorpay a large base of payment data for its AI systems. Instead of treating every payment as a separate event, the model can study wider patterns across its payment network.
Payment failure remains a major problem for online businesses. A customer may have enough money and still see a failed transaction. A bank may reject a payment, a gateway may face a technical issue, or a particular payment route may show poor performance.
Razorpay uses AI to select a payment path with a higher chance of success. Its Optimizer product studies transaction data and chooses the most suitable gateway for each payment.
Razorpay says Optimizer uses more than 2 million data points across 300 parameters and supports more than 100 payment integrations and providers. The system can handle more than 5,000 transactions per second.
Razorpay reports around a 5% increase in transaction success rates through its AI-based routing system. Its latest Vulcan deployments show a larger 8–10% improvement in payment success.
The difference matters for online businesses. A small rise in successful payments can translate into a large number of extra completed orders when a merchant handles thousands or millions of transactions.
Razorpay also uses AI to identify suspicious payments before they create losses for merchants or customers. Vulcan adds a broader intelligence layer to this work.
Early Vulcan deployments produced a sharp increase in fraud detection. Razorpay reports 8× more international-card fraud detected and stopped. It also reports 5× more fraudulent or disputed transactions identified without an increase in alerts.
The second figure has special value. A fraud system can create too many warnings if it simply marks every unusual transaction as risky. That can create extra work for merchants and risk teams. Razorpay's result suggests that its model can find more risky transactions while keeping the alert level stable.
Vulcan also supports network-level fraud detection. This approach lets Razorpay study patterns across a wider set of transactions instead of relying only on information from one payment.
Also Read - Top Payment Gateways for High-Risk E-commerce Businesses in 2026
Fraud is not the only threat. Technical problems can also cause large numbers of failed payments within a short period.
Razorpay built a real-time anomaly detection platform for this purpose. The platform handles more than 500 million transactions each month and can detect unusual payment patterns in under 30 seconds.
AWS says the platform has achieved 99.99% uptime and has reduced monitoring costs by about 80%.
The system can spot sudden changes in payment performance at the merchant and gateway level. For example, a sharp fall in payment success can trigger an alert before the problem affects a much larger number of customers.
The same system can help detect high-speed fraud patterns such as card testing, velocity abuse and unusual geographic activity. Such attacks can produce hundreds of small transactions within seconds, which makes fast detection important.
Razorpay applies AI to card authentication as well. Its ACS and Risk Engine checks customer, merchant and device information to assess transaction risk.
Razorpay says the system can achieve up to 95% authentication success, maintain 99.99% uptime and process 10,000 transactions per second.
Razorpay has also moved toward biometric and passkey-based card authentication. Its March 2026 announcement reported a 35% reduction in OTP-related authentication errors, with transaction success rates of up to 95%.
This approach tackles another source of payment failure. A genuine customer may abandon a purchase when an OTP arrives late, fails or creates extra friction. A simpler authentication path can help a legitimate transaction reach completion while the risk engine still checks for suspicious activity.
Razorpay's AI work also reaches the checkout stage. Vulcan can help decide which payment option should appear for a particular shopper.
Razorpay reports that 40% more shoppers saw their preferred UPI app through its Magic Checkout experience. The company says this change helped create an additional 100,000 to 200,000 purchases each month.
This shows a wider role for payment AI. The technology does not only react after a payment fails. It can also help create a checkout path that has a better chance of success from the start.
Razorpay also applies machine learning to e-commerce orders that use cash on delivery. Its Thirdwatch system studies more than 300 parameters and flags risky orders within milliseconds.
The system can assess signals such as address details, device information and past customer behaviour. Razorpay says Thirdwatch can reduce RTO orders by more than 60%.
For online sellers, this matters beyond payment fraud. A fraudulent or highly risky COD order can create shipping, return and handling costs even when no digital payment takes place.
Also Read - Is AI the Gateway to News Consumption Across Platforms Now?
Razorpay's AI strategy now covers much more than fraud detection. Vulcan can study billions of past payments, assess thousands of signals, select better payment paths, detect suspicious activity and support faster risk decisions.
The strongest evidence comes from the early results: 4 billion payments and 3 trillion data points used for Vulcan, around 3,000 signals per transaction, an 8–10% improvement in payment success, 8× more international-card fraud stopped, and 5× more fraudulent or disputed transactions identified without more alerts.
Razorpay is effectively turning payment data into a continuous decision system. Each transaction gives the system another signal about what works, what fails and what looks risky. That approach can help merchants protect revenue, reduce payment failures and stop fraud while keeping the checkout experience simpler for genuine customers.
The figures remain Razorpay-reported results rather than independent industry benchmarks. Still, the scale of the data and the reported improvements show how seriously Razorpay now treats AI as core payment infrastructure rather than as a separate technology feature.
1. What is Razorpay Vulcan?
Vulcan is Razorpay’s transformer-based Artificial Intelligence foundation model built specifically for payments.
2. How does Vulcan reduce payment failures?
Vulcan studies thousands of transaction signals to help identify payment routes and actions with a higher chance of success.
3. How much data did Razorpay use to train Vulcan?
Razorpay trained Vulcan on about 3 trillion data points from 4 billion payments.
4. How does Razorpay use AI against fraud?
Razorpay uses AI to identify suspicious transaction patterns and reports 8× more international-card fraud detected and stopped in early Vulcan deployments.
5. How can this help Online Businesses?
Better payment success can protect revenue, while stronger fraud detection can reduce losses and improve customer trust.
