The financial technology company Cashfree Payments is implementing artificial intelligence (AI) agents for merchants to automate payment processing. This aligns with the general trend among fintech companies in India, which are increasingly using agent scenarios to create revenue streams beyond simple payment processing.
The agent, named Relay, can be used by merchants to solve a number of tasks. These include retrying failed payments, tracking abandoned carts, confirming cash-on-delivery orders before dispatch, managing inactive subscriptions, and filing disputes.
The company aims to recover ₹20,000 crore in lost Gross Merchandise Value (GMV) through agents capable of stimulating customers who are likely to abandon a transaction via voice or text interaction, or converting them by offering discounts.
Mayank Juneja, Director of Engineering at Cashfree Payments, noted that previously, support or merchant growth teams manually contacted customers to find out the reason for abandoning a purchase or having a payment issue, offering them discounts. Now, these functions can be performed by an AI agent.
Cashfree asserts that its agent can reduce the time small and medium businesses (SMBs) spend on payment-related operations from approximately 60 hours per week to 45 minutes. Furthermore, it was stated that merchants' transactional data is not shared with external AI vendors because the agent operates on the company's infrastructure based in Bengaluru.
In addition to e-commerce, Juneja mentioned that the agent can also be applied for loan recovery purposes. This includes communicating with customers by phone, analyzing electronic payment mandates, and determining the optimal time to recontact the customer.
The launch of this service line opens a new revenue stream for fintech companies like Cashfree, which operate on low margins when processing payments in the highly competitive payments industry. Although the service is launched free of charge, Cashfree plans to introduce performance-based pricing in the future. Co-founder Ridhi Datta specified that parameters such as the number of e-commerce carts converted by the agent and successfully resolved disputes will be used as criteria for pricing as usage grows.
Meanwhile, the fintech company Razorpay has launched an AI model called Vulcan. This model is trained on four billion transactions to improve payment success rates and detect fraud. It is claimed to have been trained on 3 trillion data points and is capable of interpreting nearly 3000 signals from each transaction. A distinguishing feature of this model is that instead of specialized models for routing, risk assessment, fraud, and checkout, a single model is used that understands all the data simultaneously.