Card-Linked BNPL Seen as Fix for AI Agent Payment Failures

Traditional buy now, pay later models will break when AI agents handle purchases, according to Splitit CEO Nandan Sheth. In a new PYMNTS eBook, he argues that card-linked installment payments – which draw on existing credit rather than requiring a new application – are better suited for agentic commerce.

BNPL approval rates run between 35% and 40%, Sheth wrote. That means 6 in 10 BNPL transactions handled by an AI agent could fail at the payment step. The agent, with no human to complete a credit application, would simply move on to another merchant. The retailer would lose the sale and never know why.

“Traditional BNPL was built around a moment that stops AI agents cold,” Sheth wrote. The model requires a consumer to apply for short-term credit at checkout, receive a real-time decision and complete the transaction through a third-party provider like Affirm. That works when a human sits in front of a screen. It does not work when an agent acts on a shopper’s behalf.

Card-linked installments eliminate the new-credit application. The merchant authorizes the full purchase amount against the shopper’s existing credit card at checkout, then splits that amount into monthly payments. The credit decision is already made when the card was issued. The agent draws on available credit, not on a real-time lending decision.

Sheth called the difference critical for enterprises deploying AI agents. Businesses might only discover the BNPL limitation in production, when rearchitecting a payment integration is significantly more expensive than modeling it correctly from the start.

The card-linked model also selects for higher-quality consumers, Sheth wrote. Shoppers with established credit and available capacity tend to make larger purchases and return more often. “Card-linked BNPL is a high-quality, low-risk growth engine,” he said.

For merchants building agentic commerce, the choice of payment mechanism could determine whether agents complete a purchase or abandon it. Providers that rely on real-time credit approval – Affirm, Afterpay, Klarna – face a structural limitation in automated scenarios. Card-linked alternatives, such as those offered by Splitit, could gain share as retailers rework checkout flows.

Payment networks and card issuers stand to benefit. Card-linked installments keep the transaction on the existing card rail, generating interchange revenue and encouraging higher spend from repeat buyers. The mechanism does not require a separate loan or a third-party balance sheet.

Sheth’s argument comes as more retailers test AI agents for shopping. The eBook is part of a broader discussion on how payments infrastructure must adapt to agentic commerce, where decisions happen without human intervention.

“In the world of agentic commerce, checkout is where the hard work begins,” Sheth wrote. “Retailers who get it right will be the ones agents keep coming back to.”