As is typical with hot new developments in the payments industry, everyone is talking about agentic AI as if it appeared overnight—and everyone feels like they’re already behind the curve. The one group that doesn’t seem excited? Merchants. Much of the discussion has centered on the technology itself, not on how it will actually be deployed.
In a new report Agentic Commerce: Green Light or Flashing Yellow for Merchants?, Don Apgar, Director of Merchant Payments at Javelin Strategy & Research, explores the major questions around this emerging technology: How should a retailer apply agentic AI? How can a merchant validate that an agent is truly acting on behalf of the cardholder? And how are payment credentials secured? As Apgar notes, this much-discussed tool still has a long way to go before it matures.
The Return Quandary
Most of the work around agentic AI has focused on technology, leaving the human side largely unexamined. Take returns, for example. Merchants currently see an average of about 18% of their merchandise returned.
“If you take the human out of the equation and you have your bot making the purchase for you, what does that do to customer satisfaction and returns?” Apgar asked. “If a fifth of the stuff is coming back when you’re personally making the buying decision, how much is going to start going back when you’re not making that buying decision?”
There is also much to resolve around marketing strategies that agents will encounter. If someone is shopping for a new lawnmower, a local hardware store wants its site to appear prominently and its product details visible. But it doesn’t want a price shopping bot that only seeks the lowest price, bypassing the store’s curated merchandising strategy.
Lost Opportunities
Merchants carefully design their merchandising programs. They highlight loss leaders—like the way a Walmart or Target displays 99-cent 12 packs of soda near the entrance—knowing that shoppers are likely to add higher-margin items as they move through the store or site. With agentic commerce, that opportunity disappears. The merchant becomes a fulfillment conduit, blind to the consumer’s journey.
“When I do a search and go to a retail site, I might think: ‘Hey, this guy’s got a lot of good stuff, let me see what else he’s got. I didn’t want to spend more than $300, but maybe I will.’ Those are the kinds of human interactions you lose when the robotic agent is now doing the shopping,” Apgar said. “It’s only going by the criteria that you programmed in. It’s depriving the merchant of that opportunity to cross sell, upsell, develop a relationship, get an e-mail address.”
Indeed, many retailers have been reluctant to accommodate agentic shoppers. Ecommerce marketplace eBay, for instance, recently updated its user agreement to explicitly prohibit buyers from using AI shopping agents.
“Because it’s a marketplace and they’re not the seller, that chargeback and return problem could be massive for them,” said Apgar. “There’s nothing to say it’s going to stay that way permanently, but they’re preempting it for now until they get their arms around how they can implement this effectively.”
Four Categories with Potential
Apgar identified four retail categories where agentic AI could have the most impact. First are commoditized goods such as paper towels or dish soap, which carry little brand differentiation and are often purchased primarily based on price.
The second category is gifts. Imagine a shopper looking for a $40 gift for their nephew, a high school graduate who plays football, is a hardcore gamer, and loves blue. An AI agent could narrow options based on the criteria given and help consumers hone in on the right gift.
Third is travel, where sites like Trivago and Expedia already aggregate options. AI can streamline this process even further, reducing legwork for the consumer.
“If you are going to Chicago and want to stay down by the water on the Magnificent Mile, rather than have to go to pull up a map and see what hotels are around there, you can go to Trivago today and get that same hotel room,” said Apgar. “But if you’ve got a bot going out and doing it for you, why do you need Trivago? The bot can go right to the hotel websites.”
Finally, there are B2B applications. If a plant manager needs a widget that Grainger does not have in stock, a shopping agent could purchase it from another retailer, saving time and effort.
The Value Is Yet to Be Defined
Retailers should also note that agentic shoppers are unlikely to increase overall spending—they mainly redirect it. If a merchant invests heavily in technology to support agentic commerce, the consumer could simply buy the same item at a lower price elsewhere, reducing the store’s margin—or worse, purchasing from a competitor and leaving the merchant with no revenue at all.
“Everybody wants to be a category leader, but nobody wants to be the king of the of the low margin, high-cost sales channel,” said Apgar. “Everybody’s saying agentic AI is going to take off this year, but we’re not going to see any serious traction probably till next year. As usual, the technology is way ahead of the business.
“The technology is a facilitator,” he said. “How we get value out of the technology is yet to be defined.”
