Razorpay has launched a new artificial intelligence model built only for payments. The fintech company calls the model Vulcan. It aims to make digital payments more reliable, reduce fraud and help more customers complete online purchases.
Razorpay built Vulcan with NVIDIA and AWS. The model has been trained on nearly 3 trillion data points across 4 billion payments. It also studies about 3,000 signals for each transaction before it makes a decision.
The launch comes at a time when digital payments have become a normal part of daily life. Yet, a payment can still fail for many reasons. A bank may reject a transaction, a card may not work with a certain route, a customer may see the wrong payment option, or a transaction may look risky.
Razorpay wants one AI system to deal with these issues instead of using separate tools for each problem.
What Is Vulcan?
Vulcan is an AI foundation model made for payment data. Unlike a normal large language model, which learns from text, Vulcan focuses on patterns in financial transactions.
The model looks at many details from a payment before it makes a choice. These details can include the payment method, bank, card, transaction value and other signals linked to the payment process.
The main aim is simple. Vulcan tries to find the best way to complete a payment while also checking for signs of fraud or risk.
Razorpay says the model acts as one intelligence layer across payment routing, fraud, risk and checkout personalisation. This is different from the older approach, where a company may use one machine-learning model for payment success, another for fraud and another for customer choice.
Nearly 3 Trillion Data Points
The size of the data used for Vulcan is one of the biggest parts of this announcement.
Razorpay says the model has learned from nearly 3 trillion data points across 4 billion payments. For each transaction, it can use about 3,000 signals.
This gives the model a large base from which it can learn payment patterns. A payment may look simple to a customer, but many small factors can affect its result.
For example, one bank may work well for a certain card at one time but show a poor success rate at another time. A specific payment route may also work better for one type of transaction than another.
Vulcan can study such patterns and make a choice based on the data available at that moment.
The model can make its decision in about 29 milliseconds, according to Razorpay. That speed matters because customers do not want to wait several seconds while a payment system decides what to do.
Better Payment Success
One of the biggest goals of Vulcan is to reduce failed payments.
Razorpay says early parts of the model have already shown an 8% to 10% improvement in payment success rates. That could make a major difference for online businesses because a failed payment can cause a customer to leave without completing a purchase.
For a shopper, the problem may look very small. They may simply see a message that says the payment failed. But for a business, repeated failures can mean lost sales.
Vulcan can assess different payment routes and choose the route that has a better chance of success. This can happen in real time, without asking the customer to make the choice.
The system can also recommend which saved payment method is more likely to work for a particular customer. This could make the checkout process easier and reduce the number of failed attempts.
Stronger Fraud Detection
Fraud is another major focus of Vulcan.
Razorpay says early tests showed that the system detected eight times more international card fraud. It also identified five times more fraudulent or disputed transactions without an increase in the number of alerts.
This point is important because too many fraud alerts can create a new problem.
If a payment system flags too many normal transactions, genuine customers may face extra checks. Businesses may also lose sales because a safe payment gets treated as risky.
Razorpay says Vulcan can find more risky activity without simply sending more alerts. The goal is better detection rather than a larger number of warnings.
The model can also identify risky Cash on Delivery orders, which adds another use for the system beyond standard digital payments.
A Better Checkout Experience
Razorpay also plans to use Vulcan to improve checkout.
Its Magic Checkout product helps customers choose a payment method when they buy something online. Razorpay says 40% more shoppers are now seeing their preferred UPI app through Magic Checkout.
The company says this has helped complete an additional 1 lakh to 2 lakh purchases every month.
This may seem like a small change, but payment choice can have a direct effect on sales.
A customer may already prefer one UPI app and feel more comfortable with it. If that app does not appear at checkout, the customer may need to search for another option. Some people may even leave the purchase.
Vulcan can use payment patterns to help show the option that is more likely to work for that person.
Tests Across 1.5 Million Shoppers
Razorpay says early parts of Vulcan have already been live for a few weeks.
The system has been tested across about 1.5 million shoppers and more than 51,000 businesses. Companies such as Blinkit, Bachatt and redBus are among the customers that have used the technology in live payment environments.
This gives Razorpay a chance to test the model under real conditions instead of only inside a laboratory.
Real payment data can be complex. Customers use different banks, cards, UPI apps and devices. Their payment habits can also differ based on location, purchase value and time.
A model that works well across these different cases could have a strong advantage.
Built With NVIDIA and AWS
Razorpay built Vulcan with support from NVIDIA and AWS.
NVIDIA provided compute and architecture expertise, while AWS provided its cloud platform and scaling expertise, according to Razorpay co-founder and CEO Harshil Mathur.
The company says the model was trained on Razorpay’s own payment data and operates within Razorpay’s infrastructure. Mathur also said payment data was not shared with third parties.
This is important because payment information is highly sensitive. Businesses and customers need strong controls over financial data.
Keeping the model within Razorpay’s infrastructure also gives the company more control over how the system uses its data.
Why Razorpay Chose a Foundation Model
Razorpay’s larger idea is not limited to the four areas where Vulcan works today.
The company says the foundation model can support more use cases later. These may include marketing and lending, depending on the results of future tests.
That gives Vulcan a much wider role inside Razorpay.
Instead of building a new AI system each time the company wants to solve a new problem, it can use the same core model and apply it to another area.
This could save time and help Razorpay create new products faster. However, each new use will still need careful tests because financial decisions can have serious effects on customers and businesses.
What This Means for Online Businesses
For merchants, better payment performance can have a direct effect on sales.
If more customers complete their payments on the first attempt, businesses can avoid some lost orders. Better fraud checks can also help reduce losses from risky transactions.
A system that works across payment routing, fraud and checkout could make the entire payment process more connected.
That is the main idea behind Vulcan. Rather than treat each part of a transaction as a separate problem, Razorpay wants its AI to look at the full payment journey.
Razorpay’s Bigger AI Plan
The Vulcan launch also shows that Razorpay sees AI as a major part of its future.
The company has a large amount of payment data from its network. That gives it a useful base for models made for financial transactions.
The challenge is to turn that data into systems that create clear value without harming trust or security.
So far, the early numbers give Razorpay a strong start. An 8% to 10% rise in payment success, eight times more international card fraud detection, five times more fraudulent or disputed transaction detection, and 40% more shoppers who see their preferred UPI app are significant results.
The Road Ahead
Vulcan marks a major step in Razorpay’s AI plans. The model combines nearly 3 trillion data points, 4 billion payments and about 3,000 signals per transaction to make fast decisions.
Its first focus is clear: help payments work better, detect more fraud and make checkout easier.
The bigger opportunity may come later. If Razorpay can use the same foundation for areas such as lending and marketing, Vulcan could become a core part of its wider business.
For now, the company has shown that AI can do more than create text or answer questions. In Razorpay’s case, it can study millions of payment patterns and make a decision in about 29 milliseconds.
That could make the simple act of paying online faster, safer and more reliable for millions of people and businesses.
Also Read – Copy These AI Startup Ideas Before They Go Mainstream