Payments technology company Razorpay has introduced an artificial intelligence model named Vulcan, designed to enhance payment success rates and detect fraud. The Bengaluru-based firm developed this proprietary model, which utilizes data from four billion transactions and incorporates three trillion data points, enabling it to analyze nearly 3,000 signals from each transaction.
Unlike previous systems that used specialized models for different functions like routing or fraud detection, Vulcan employs a unified approach, processing all data points simultaneously. CEO and founder Harshil Mathur noted that this AI-driven model continually learns from each transaction, improving with every interaction.
Vulcan’s architecture is developed in collaboration with experts from Nvidia and Amazon Web Services (AWS) and is capable of various functions, including routing, network-level fraud detection, risk assessment for cash-on-delivery transactions, and predictive personalization at checkout.
A study conducted by Razorpay involving 1.5 million shoppers and over 51,000 businesses indicated that payment-related issues such as failed transactions and delays are prevalent across different market types, from urban to rural areas. The model has reportedly improved transaction success rates by 8% to 10% and enabled 40% more shoppers to locate their preferred Unified Payments Interface (UPI) apps at checkout, which results in an increase of 100,000 to 200,000 successful purchases each month. Additionally, the model has improved fraud detection, identifying eight times more international card fraud and five times more fraudulent transactions without increasing alert volume.
Why this story matters
Key takeaway
Opposing viewpoint
