Agentic AI will shake up charge-backs. What banks need to know

• Key insights: As agentic commerce takes off, charge-offs will initially spike, then start to ease.
• What's at stake: Managing payment disputes is still mostly done internally, creating an opportunity for outside providers as technology becomes more complicated.
• Expert quote: "It's going to be messy for the next five years," said Armen Najarian, chief marketing officer at Sift,
  

With agentic commerce gaining steam, some payments professionals anticipate a jump, then an eventual drop in charge-backs, a pervasive and expensive problem for banks.

Globally, Datos Insights predicts that charge-back volume will climb 24% from 2025 to 2028 — reaching 324 million globally. Merchants and issuers state their charge-back volume has increased more than 10% in the past year. That's according to a March report by Ethoca, a Mastercard company, using data from Datos Insights.

Dealing with charge-backs is costly and labor-intensive. U.S. financial institutions require one full-time employee for every $13,000 to $14,000 in incoming annual cardholder disputes, and each dispute costs them between $9.08 and $10.32 to process, according to the report. 

Agentic commerce, however, has the potential to decrease charge-backs, which payments professionals predict could happen eventually after some of the initial wrinkles are ironed out. "It's going to be messy for the next five years," Armen Najarian, chief marketing officer at Sift, an AI-powered fraud decisioning company based in San Francisco, told American Banker.

Here's what banks need to know about charge-backs as the shift to agentic commerce gains prominence:

The potential to reduce friendly fraud

Many charge-backs are considered friendly fraud. This means a customer, intentionally or mistakenly, disputes a legitimate transaction instead of seeking a refund directly from a merchant. With agentic commerce, there's strong potential for this type of misuse to be significantly reduced, Najarian said. 

That's because, as agentic commerce matures, there will be clear definitions regarding how agents are authorized and approved, which limits the ability of consumers to say they didn't make a purchase. With agentic commerce, "you've really got to work hard to prove it wasn't you," Craig DeWitt, co-founder and head of product at Skyfire, a San Francisco-based payments fintech, told American Banker.

Even so, Alisdair Faulkner, the chief executive and co-founder of Darwinium, a cyberfraud AI platform with offices in San Francisco, London and Sydney, said friendly fraud claims are likely to increase initially due to genuine consumer mistakes using a new purchasing method. This should decrease in time, he said. 

Reducing fraud by bad actors

One of the challenges being addressed is how merchants can recognize a good bot from a bad bot. A lot of technology today is aimed at blocking bots from making fraudulent purchases during the checkout process, but in the age of agentic commerce, changes have to be made so that legitimate bots can complete the checkout experience, said DeWitt of Skyfire, whose technology allows agents to legitimize themselves for this purpose. 

Especially in the early days of agentic commerce, bad actors will try to flood the system to see where they are caught and where they are not, said Elias Ghanem, global head of Capgemini Research Institute for Financial Services. This could mean more attacks get through at the onset, but eventually, he said, things should become more manageable as more barriers are erected to keep the bad guys at bay and reduce charge-backs due to fraud. But it can never be 100% secure, Capgemini's Ghanem told American Banker. "It's a never-ending race between the honest and the dishonest people. There's so much money to be stolen, unfortunately, that they will continue to have an incentive to do it."

Establishing protocols

To be sure, obtaining a meaningful drop-off in charge-offs due to agentic commerce could be years away. That's because there's a lot that has to happen behind the scenes so that agentic commerce can take off. For starters, the card networks are establishing standards to define the technical framework for agents and proving that a user gave an agent specific authority to make a particular purchase. These protocols are also meant to help merchants ensure that an agent's request accurately reflects the user's true intent and determines accountability if fraud or an incorrect transaction occurs.

Visa announced initial specifications for its Trusted Agent Protocol in mid-October. Separately, Google announced an Agent Payments Protocol in September, along with a coalition that includes more than 60 organizations, including Adyen, American Express, Ant International, Coinbase, Etsy, Forter, Intuit, JCB, Mastercard, Mysten Labs, PayPal, Revolut, Salesforce, ServiceNow, UnionPay International and Worldpay.

Challenges remain

Beyond standard-setting, consumers have to be willing to use agentic commerce more readily than they are now, Sift's Najarian told American Banker. Only 14% of respondents said they would let an AI shopping agent make purchases on their behalf, according to a recent Sift survey of 1,000 consumers. Najarian predicts this will proliferate over time. As education becomes stronger, more consumers will be willing to try using agents, he said.

Signs of growth

The growth of agentic commerce also turns on merchants embracing the new model. This is likely to start with larger merchants and trickle down to smaller ones, payments professionals said. Just recently, Walmart announced a partnership with ChatGPT to allow customers to complete purchases from the retailing giant directly within the AI agent.

In another notable move that suggests agentic commerce is gaining traction, Mastercard said last month that its agentic payments platform, Mastercard Agent Pay, would be integrated into PayPal's wallet to allow AI agents to simply and securely complete transactions on behalf of PayPal users.