Cashfree launches AI agent Relay

Fintech firm Cashfree Payments is rolling out artificial intelligence agents to merchants. The move aims to automate payment operations and open new revenue streams beyond basic processing.

The AI agent, named Relay, is designed to handle several key tasks. According to the company, these include retrying failed payments, following up on abandoned shopping carts, and confirming cash-on-delivery orders before dispatch. It also manages failed subscriptions and files payment disputes.

Mayank Juneja, director of engineering at Cashfree Payments, explained the shift. He stated that functions traditionally performed by human growth teams, such as calling customers about abandoned carts or offering discounts, will now be powered by the AI agent.

Cashfree is targeting a significant amount of lost sales volume. The company aims to recover Rs 20,000 crore in gross merchandise value (GMV) currently lost to abandoned transactions. The AI agent can nudge customers likely to abandon a purchase through voice or text interactions, sometimes by offering discounts to complete the sale.

Operational Impact for Merchants

The company claims the agent can drastically reduce the time merchants spend on payment-related tasks. A small or medium business might spend around 60 hours a week on these operations. Cashfree states its AI agent can cut this down to approximately 45 minutes.

Relay operates on Cashfree's own infrastructure. The company emphasized in a statement that merchant transaction data is not sent to external AI providers, addressing potential data privacy concerns.

Broader Applications and Market Context

The application of Relay extends beyond standard e-commerce. Juneja noted the agent could also be deployed for use cases like loan recovery. This would involve following up with customers over calls, examining e-payment mandates, and determining optimal times for re-contact.

This launch represents a strategic expansion for fintech firms operating on thin margins. Cashfree said the service is free at launch but plans to introduce an outcome-based pricing model in the future. Cofounder Reeju Datta indicated that metrics like the number of converted shopping carts or successfully resolved disputes would serve as pricing parameters as adoption grows.

The move follows similar industry developments. This month, rival fintech firm Razorpay rolled out an AI model named Vulcan, trained across four billion transactions. Razorpay claims its model was trained on 3 trillion data points and can understand nearly 3,000 signals from each transaction to improve payment success and detect fraud. The company highlighted that Vulcan is a single model handling routing, risk, fraud, and checkout, instead of using separate specialized models for each function.