Are Retailers Ready for AI to Foot the Bill? How Payments Infrastructure Must Adapt for Agentic Commerce

Artificial intelligence is helping shoppers find their next favorite products, but will they start letting it check out too? That's looking more and more likely, as over half of U.S. consumers are using AI at least occasionally to help make purchasing decisions, and now 39 percent say they would let it make authorized purchases on their behalf.

For retailers watching this trend unfold, it’s easy to see that AI could help remove friction between product discovery and checkout, reducing abandoned carts and bringing in extra revenue.

However, retailers will need to work closely with their payment providers to verify that an agent is legitimate, confirm what the customer authorized, and retain evidence of that action. With 48 percent of consumers concerned about privacy and data security as AI enters financial services, any uncertainty could lead to disputed transactions and greater hesitation among shoppers.

Understanding Customer Intent and Maintaining Trust

With an agent-led purchase, the customer wants to be hands off and may never see the checkout page, even though the payment comes from their account. The retailer and payment provider must ensure they can verify who an agent represents and what the customer has authorized. Once those checks are complete, the payment provider can process and settle the transaction through blockchain or existing payment rails.

For instance, a customer might authorize an agent to buy a large winter coat from an approved retailer, spending no more than $300, and in their favorite color. More subjective notes like favorite color may cause challenges for traditional rules-based systems, so agentic models will need to incorporate upfront clarifying questions, such as to confirm which color. This digital mandate helps capture a customer’s intent. Retailers and their payment providers can then compare the agent’s behavior with those instructions as the transaction unfolds.

If a purchase exceeds the spending limit or falls outside the mandate, it may need to be stopped or sent to the customer for confirmation. Those decisions must happen quickly enough to protect the customer without undermining the convenient checkout experience agentic commerce provides.

Establishing Responsibility When Purchases Go Wrong

Even with real-time controls, some agent-led purchases will go wrong. The transaction record must then show what the customer authorized, how the agent behaved, and where the failure occurred.

If inaccurate product information from a merchant leads an agent to make the wrong purchase, liability may sit with the merchant. If the agent exceeds the customer’s budget, the agent operator may be responsible. Retailers should ask their payment partners how agent-led disputes will be investigated, what evidence will be available, and how liability will be assigned.

Preparing for Competing Standards

OpenAI’s Agentic Commerce Protocol and Google’s Universal Commerce Protocol are early attempts to govern how agents discover inventory and receive limited access to payment methods. Visa and Mastercard are also developing approaches to agent authentication.

There's not yet a universal credential that allows an agent to prove its authority across merchants, therefore retailers should look for payment partners capable of supporting several protocols.

For shoppers, the promise of agentic commerce is a purchase that happens almost invisibly. But behind the scenes, every step needs to be traceable: what the customer authorized, what the agent did, and what happened if it went beyond those instructions. Retailers and payments providers that that build this accountability into their payment infrastructure can adapt as protocols evolve and give customers the confidence to let AI purchase on their behalf.

Kaue Tozzi is head of product strategy at Temenos, a cloud-native, AI-driven banking software.