Fintech firm Cashfree Payments is rolling out artificial intelligence (AI) agents to merchants to automate payment operations, as fintech firms in India increasingly turn to agentic use cases to build revenue streams beyond payment processing.
The AI agent, called Relay, can be deployed by merchants for use cases such as retrying failed payments, following up on abandoned carts, confirming cash-on-delivery orders before dispatch, managing failed subscriptions and filing disputes, among other things.
The company is targeting ₹20,000 crore in lost gross merchandise value (GMV) through agents that can nudge customers who are likely to abandon a transaction by initiating a voice- or text-based interaction, or convert them by offering discounts.
“Today, merchants have an ops or growth team who would call customers to understand why they would have abandoned their cart or if they faced issues during a payment. These teams could give a customer a discount. These functions would now be powered by an AI agent,” said Mayank Juneja, director, engineering, Cashfree Payments.
A small and medium business (SMB) could spend around 60 hours a week on operations related to payments. Cashfree claims to cut this time to 45 minutes with its agent.
The agent runs on the Bengaluru-based company’s infrastructure. “Merchant transaction data is not sent to external AI providers,” it said in a statement.
Beyond e-commerce, Juneja said the agent could also be deployed for use cases such as loan recovery. This could include following up with customers over calls, examining e-payment mandates, and determining the right time to contact a customer again.
The launch of the agentic service opens up a new revenue line for fintechs like Cashfree that operate on thin payment-processing margins in the competitive payments industry.
Cashfree said while the service is free at launch, it plans to introduce outcome-based pricing in the future.
Co-founder Reeju Datta said outcome-based pricing would mean that metrics such as the number of e-commerce carts converted using the agent and disputes successfully resolved would serve as pricing parameters as adoption grows.
This month, fintech firm Razorpay rolled out an AI model, Vulcan, that is trained across four billion transactions to improve payment success rates and detect fraud.
It claims to have been trained across 3 trillion data points, with the model able to understand nearly 3,000 signals from each transaction. What is different in this model is that instead of a specialised model each for routing, risk, fraud and checkout, a single model like this understands all data points at one point.
