Razorpay launches Vulcan, an innovative AI foundation model in India designed to enhance payment success, reduce fraud, and personalise checkout experiences, developed with Nvidia and AWS technology.
Razorpay has unveiled Vulcan, which it describes as India’s first artificial intelligence foundation model built for payments, in a bid to make digital transactions more reliable, safer and easier to complete. The company says the system was developed with Nvidia and AWS technology and trained on nearly 4 trillion data points drawn from 4 billion payments, giving it a wide view of how payment infrastructure behaves across merchants and conditions.
The model is designed to tackle several separate problems at once. Razorpay says Vulcan can improve payment routing, detect fraud, assess return-to-origin risk and personalise checkout choices. In its launch material, the company said early testing showed an 8% to 10% rise in payment success rates and an eightfold improvement in spotting international card fraud. Business Standard reported that the model processes around 3,000 signals per transaction, which Razorpay argues allows it to make sharper decisions in real time.
At the heart of the system is a simple idea: payments are not all the same, even when they look similar to shoppers. A card, UPI payment, wallet or netbanking transaction can behave differently depending on the bank, provider, merchant setup and time of day. Razorpay says Vulcan was built to learn from those differences at scale, rather than rely only on fixed rules that can become stale or on retries that arrive too late to improve the first attempt.
The company also says the model is intended to use network-level patterns that individual merchants cannot easily see on their own. That matters for fraud, where attacks can be spread across many businesses in ways that look ordinary in isolation, and for checkout ranking, where the system must infer which option a shopper is most likely to complete. Razorpay says the model does not decide what a merchant is permitted to offer; instead, it ranks choices that are already allowed.
Razorpay’s own blog says the model was built around a stricter measurement framework than a simple before-and-after comparison. It describes out-of-time testing, flow-by-flow checks and fixed alert thresholds for fraud, all meant to reduce the risk of overstating results. The company also says direct identifiers such as names, email addresses, phone numbers and bank account numbers are excluded from the training pipeline before data is used to build the model.
Disclaimer: This article is intended to inform and educate, not to recommend or endorse any financial product, investment or strategy. Please consider your own financial circumstances and seek professional advice where appropriate before making financial decisions.
