Although generative artificial intelligence has powerful proficiencies, its limitations have become more apparent as the technology is deployed at scale. Now, however, Visa and Mastercard have unveiled platforms aimed at giving AI an even more prominent role—AI agents will soon be able to shop and make purchases with minimal consumer input.
In a recent PaymentsJournal podcast, Christopher Miller, Lead Emerging Payments Analyst, Suzanne Sando, Lead Fraud Management Analyst, Don Apgar, Director of Merchant Payments, Jordan Hirschfield, Director of Prepaid, and James Wester, Co-Head of Payments at Javelin Strategy & Research, discussed the barriers to agentic commerce, potential use cases for the emerging technology, and the broader future of AI.
A Vision of the Future
In both Mastercard’s Agent Pay and Visa’s Intelligent Commerce platforms, AI agents are effectively taking on the role of personal shopper.
“In this vision, the agent is everywhere,” Miller said. “It’s making your life easier; it’s saving you time; it’s relieving you of the burdens of your side of any transaction. It can find items for you to purchase, it can choose merchants for you to purchase from, and it can select which form of payment you wish to use at any given point in time. There’s a lot behind that vision, and many technical aspects will have to be addressed for a system like that to operate.”
Beyond concerns about security or accuracy, one of the main barriers to agentic commerce is the need for active customer participation. Until now, organizations have largely implemented AI initiatives behind the scenes.
Persuading consumers to entrust their payment data to an AI agent will likely require a deeper understanding of AI than many users currently have.
“In the product discussion that Visa had when they did their product launch talking about agentic, one of the things that resonated with me was it was one of the products where people said, ‘We are going to have to pull consumers along,’” Wester said. “’We are going to have to show them and educate them on the magic of agentic commerce.’”
In addition to the awareness obstacle, there is growing skepticism about the accuracy of AI-produced information. While many concerns about artificial intelligence are unfounded, there are documented limitations that could cause consumers to be wary of agentic commerce.
“If you’ve used AI, the answers that it returns are just as often wrong as they are right,” Apgar said. “When you give a shopping agent a test to go find something and buy it for me, it’s not the buying part that worries me, it’s the finding the right thing. In today’s environment, you’ve also got AI hallucinations. It’s not everybody’s worst fear, but if you get a message that says, ‘Hey, I’ve bought this for you’ it’s like, ‘Wait, who asked for that?’”
The Potential for Exploitation
Along with concerns that an AI agent could make a costly mistake, it is inevitable that bad actors will attempt to exploit or manipulate these systems.
On one hand, fraud prevention has been a standout use case for AI, as it can detect suspicious activity across vast amounts of data. On the other hand, the emergence of a new technology connecting consumers, merchants, and financial institutions introduces risks that must be addressed.
“There’s obviously a major need for continuous authentication throughout the entire life cycle of the agent interaction,” Sando said. “At this point, we’re thinking data points, what behavioral analytics can I use to continuously authenticate the consumer to the agent? This is just like any other identity verification and continuous auth scenario.”
“It doesn’t matter what the channel is, whether it’s your digital wallet or you’re going straight to a merchant or you’re using the actual agent, even down to when you’re planning for situations where social engineering is suspected,” she said. “The best scam detection solutions are looking for those real-time clues where social engineering is taking place where a criminal is convincing a consumer to make a certain purchase.”
Many organizations use behavioral analytics to identify such clues—for example, when a user makes a purchase that is out of character or from a significantly different location.
However, if it’s unclear to a retailer whether a purchase is made by an AI agent or a human, it can be difficult for the business to distinguish fraudulent activity. In this scenario, merchants must rely on the agent service to have conducted proper due diligence in detecting illegitimate or unauthorized behavior before initiating the purchase.
Additionally, the services themselves present a potential avenue for cybercriminals exploitation.
“We should be expecting situations where criminals are creating fake websites and apps that offer a similar service,” Sando said. “They’re going to try and convince consumers to sign up for what they think is a legitimate agent service and then in turn, they will be giving up a whole host of PII and payment information and data for this particular scam.”
“On top of that, we should be expecting a surge in text and email scams from fraudsters that are impersonating legitimate agent services,” she said. “Not only do we have to worry about fake services, but now we’re worried about the use of generative AI that has already made impersonation scams easy for criminals to commit. I don’t think it’s at all far-fetched to assume that agentic commerce will be affected as well.”
The Training Wheels
While significant concerns surround agentic commerce, there are also powerful benefits. Many consumers struggle to fully leverage loyalty points, gift cards, and other rewards. In these split-tender scenarios, an AI agent could dramatically impact the customer experience.
“If you can split-tender, I can get the most value for my money by saying it’s going to understand how it can redeem reward points in a stored value account,” Hirschfield said. “How can I benefit if it’s earning more reward points through a particular card that it’s going to use? How does it pick and choose how to split up a payment to buy something that most benefits the consumer, so they get the most value for their money?”
Loyalty and rewards have become integral to prepaid accounts, which may also serve as the bridge connecting consumers to agentic commerce.
“Prepaid can be the training wheels,” Hirschfield said. “Obviously, you need to train the models and you need to train the consumer. With prepaid, consumers are already used to turning money over to prepaid accounts and stored value accounts.”
“Part of the issue is how do I trust a third party—be it the retailer themselves with a stored value account or an agent—with my money,” he said. “If I limit how much I give them and I’m giving them a pot of money, this is a more direct way to program that money.”
Agentic AI could also offer potential benefits for merchants. If a retailer can recognize that an AI agent is making a purchase, it may open the door to a new dynamic in commerce.
“Can the merchant now offer agent incentives to buy at their store versus somebody else’s store?” Apgar said. “If Target and Walmart have the same item and Target is now paying a sales performance incentive fund to the agent to make that purchase at Target versus making it at Walmart, now the merchant has the ability to apply leverage to the automated shopping process.”
The Limits of Imagination
While the impacts on consumers and merchants are yet to be determined, the promise of agentic AI suggests that an entire industry may emerge to power this new paradigm.
“At the end of the day, to the extent that this vision comes to pass, it is the providers of the agents themselves who are most likely to insert themselves as a new layer of the commerce stack and extract some form of value,” Miller said. “We’ve seen this play out repeatedly, so I’m going to suspect that’s who will benefit the most. Whether other participants benefit or not, I think is a TBD.”
Although many questions remain about the trajectory of agentic commerce, there is little doubt that it is gaining momentum. The launches of Visa and Mastercard’s platforms are imminent, and they are likely just the first of many agentic AI-powered initiatives.
“Skepticism is warranted, but this is happening,” Wester said. “If we are saying, ‘I can’t imagine why somebody would do something,’ that shows the limits of our imagination, not the limits of where this is going to go.”
“Approaching this with an open mind and understanding that there is going to be an entire industry of developers, systems integrators, and folks that are going to be aimed at this (if important),” he said. “It’s understanding that this is bigger and important, and we need to understand that in the context of our entire industry, as opposed to just saying this seems like a lot of hype.”
The post The Agentic Advent: How the Next Iteration of AI is Shaping Commerce appeared first on PaymentsJournal.
