Why AI Shopping Agents Are Retail's New Mall Anchor

Every morning, they start showing up to shop. By the time the East Coast is at work, the crowd is in the tens of millions.

They come with an occasion. A wedding in September on the Cape. A baby shower gift for a friend’s daughter. Back-to-school clothes and supplies. The It bag for date night in two weeks.

Some know exactly what they want and head straight for it. More have a rough idea. Blue, not gray. Sleeveless, not long sleeves. Tote versus top handle bag. They need help getting to a short list. So, they look around. They compare a few things, read the reviews, check the prices. They see a lot and narrow it to three.

It’s the busiest shopping day of the week. And it’s a Tuesday.

There’s no parking lot. No food court either. No anchor at either end, no long walk down the corridor between them, no opening or closing hours. Nobody does ten thousand steps doing double duty while shopping.

The mall with the parking lot is having a very different morning.

Coresight Research counted 8,892 U.S. store closings in 2025 and expects another 7,900 this year. Green Street, which grades more than a thousand American malls, ties at least 175 mall closures over the past 15 years to the unwinding of the department store model. And the split inside what’s left is getting dicey. Coresight found that foot traffic at non-top-tier malls fell 1.6% in 2025, the first decline since the pandemic, while the top tier grew 2%.

Two malls. Two mornings. Two outcomes. Except the mall that’s winning right now has a problem the mall that’s dying solved sixty years ago.

The shopper who shows up and browses there leaves and buys somewhere else. That’s the part that makes every merchant reading this feel pretty happy with the status quo.

It shouldn’t.

The Scariest Number in Retail: Why AI Referral Traffic Converts 54% Better

Adobe Analytics, looking across more than a trillion visits to U.S. retail sites, found that in May 2026 shoppers arriving from an AI referral converted 54% better than shoppers arriving from anywhere else. They spent 53% more time on the site and browsed 23% more pages. A year earlier, they reported that same traffic converted worse than the rest of their traffic.

If you run a store, that number gets a lot of high-fives. Somebody else helped with the browsing, the comparing and the persuading, then sent the shopper to you with her mind made up and her card out. No acquisition cost. No discount. She converts better than the customer whose CAC cost you a lot. What’s not to love?

Retailers probably think this is the best deal going. The agent is a referral source. A very good one. Why connect to the models and risk losing control of the customer relationship and my business model when they’re a new affiliate channel that costs me exactly zippo right now? And exactly zero sales are happening now there anyway.

I want to persuade every retailer that this is the riskiest assumption they can make right now. The number that makes you sit back comfortably and count the dollar bills like manna from heaven is the same number that should make you sit up in the middle of the night sweating bullets.

What’s delivering that shopper to retailers right now isn’t simply a free referral source. It’s an agent playing the role of anchor store with a mall attached to it that has everything except a cash register. And the entire question that should be front and center for every retailer in America is what happens on the day one of them installs one.

Andy Grove’s 10X Force: What Intel’s Missed iPhone Bet Teaches Retail About AI

Andy Grove wrote Only the Paranoid Survive in 1996. The idea at the center of it seems almost palm-to-forehead obvious. But like a lot of things in business, the simple part is mostly what you see when looking in the rearview mirror after something big happens in the business.

In the book, Grove said the forces acting on a business don’t hold still. Competitors, suppliers, customers, new entrants, substitutes, complementors are all changing, moving in response to market dynamics. Occasionally one of them gets an order of magnitude more powerful than what the business is built to handle. There’s wind and then there’s a typhoon, he wrote. There are waves and then there’s a tsunami. He called it a 10X force, and the moment the underlying structure begins to show its cracks, a strategic inflection point.

At Intel, Grove saw one coming. He looked at the internet and understood exactly what it was going to do, and Intel spent the next decade winning it. He was as right as an executive gets to be.

And then he and his company missed the smartphone.

Not the phone itself, but what the phone turned out to be. The small internet appliance that would end up carrying the interest, the attention and eventually the commerce. And the chips. Intel had a chance at it and passed. Paul Otellini, Grove’s successor twice removed, said years later that Apple wanted a chip at a price below Intel’s forecasted cost. “I couldn’t see it,” he said. “It wasn’t one of these things you can make up on volume.” In hindsight, he recounted, the forecasted cost was wrong, and the volume was 100x what anyone thought.

It’s a story about how a good executive misjudged the 10X force when it actually arrived. That’s because it doesn’t show up at the door in its Sunday finest and announce itself as a threat. It shows up looking small, low-margin, irrelevant, even friendly. Insignificant, playing the role of something that couldn’t possibly matter to a business as large and as correct as the one you’re running today.

Retail’s AI Shift Looks Just Like the Ecommerce Boom of 1999

Retail got its 10X force in 1999, and it was called eCommerce. It took retailers a lot of time to lean in. When they did, they built the sites, the apps, the marketplaces, the buy-online-pick-up-in-store, the loyalty file, the omnichannel mission statement and playbook.

So, retail knows exactly what a 10X force feels like. That’s part of the problem. It’s now organized, staffed, budgeted and incentivized around the last one. And sort of reveling in the fact that 80% or more of retail sales still happen in the physical store.

That explains why the next 10X force looks a little harmless to many retailers, at least right now. Ignoring the reality that in some categories, like apparel and beauty supplies and sporting goods and electronics, Amazon has built a retail juggernaut that captures 16%, 7%, 35% and 32% of retail sales, according to PYMNTS Intelligence.

And the shopping habits that are being formed when consumers show up at the prompt.

The 2026 Global Digital Shopping Index from PYMNTS Intelligence, in collaboration with Visa, studied 5,841 consumers and 1,185 merchants across the United States, Brazil and the United Arab Emirates. It found that 47% of online shoppers used AI somewhere in their most recent purchase journey, and that the share using ChatGPT for product search went from 2% to 30% in two years.

That’s a 15X increase in two years. A shift that’s already happened.

Adobe’s numbers would say that today the agent is the best complementor a merchant’s ever had. It finds the shopper, it qualifies her, it persuades her and it hands her to the retailer. For free.

Until one day soon when it may not do either.

How AI Shopping Agents Became Both the Anchor Store and the Mall

The physical mall ran two different retail plays. The anchor gave a shopper a reason to make the trip. Nordstrom or Macy’s was why she got in the car on a Saturday, and the mall operator knew it. That’s why the anchor got its space for almost nothing.

The mall had its own playbook. It took the traffic the anchor created and put forty stores in the path of it. Most of them were not really the draw. The shopper showing up with latte in hand had never even heard of half of them.

One type of store created the demand, the reason for the trip. The other monetized the walk between the anchors that were the demand generators, hoping to catch some sales.

In an agentic world, the agent can do both. In the same prompt box. The prompt is the anchor. It’s the reason the shopper shows up at all. What returns in her feed is the mall. A set of stores assembled for the occasion she asked the agent to go hunt down. Arranged in an order the agent chose, that didn’t exist before she typed and won’t exist after she’s done.

That’s what makes the agentic commerce shift so different from the other retail transformations retail has navigated. Search was a directory. Marketplaces were a bigger store. Agents become the anchor and the landlord all rolled into one.

In the physical mall they were separate parties with separate financial interests and incentives. The anchor could threaten to leave. The mall operator could go court a different one. A tenant could play them against each other and sometimes win. Or deck out the window with the fanciest displays to court shoppers. Or give away free stuff to get them inside.

In the agentic world there’s no breathtaking window displays. The thing that creates the traffic and the thing that decide who gets to stand in front of it are the same things. It rebuilds the virtual tenant mix for every single prompt. Over time, it will change the business model and economics of the shopping experience for the merchants who show up. Anyone who’s ever bought a search ad to appear above their own organic listing already knows that the change feels like, and what it costs them.

The Hidden Risk of Free AI Referral Traffic for Retailers

Retailers reading the Adobe stats might still wave their hands and say, “not my worry and not right now. There are absolutely zero transactions happening on LLM models, so I’m not losing sales to the competition. And, I’m getting the sale anyway. So come back later when there’s something to talk about.”

Here are four reasons why that logic is bound to go sideways.

Sure, you got the transaction. But you lost the intent and the ability to shape the buy. That 54% is high because the selling already happened somewhere when you weren’t in the room. Retail’s margin has never been in taking the money. It’s in being the reason she chose you. Strip that out and what’s left is a store that still pays rent, still pays for the site, still pays for the brand campaign, and now competes on the only variables the agent can read. Price, stock and delivery date. So, congratulations, you’ve been promoted to fulfillment.

Also, you can’t measure the sales you didn’t get. You can count every order that comes tagged with an AI referral. You can’t count the times you weren’t on the list. There’s no line in your P&L for the consideration set you didn’t make, no bounce rate for a page she never saw. That 54% is a survivorship number measured entirely on the shoppers who were sent to you. The dashboard will look better and better. Right up until the moment it doesn’t. By then you’ll be left wondering why the sales no longer happen.

Then there’s the no-such-thing-as-a-free-lunch part. Every one of those prompts teaches the agent something. What the shopper wrote at the prompt, what she picks, what she rejects, what price cleared, which delivery promise closed it, what came back and why. Retailers only learn that she showed up. That data imbalance compounds daily with each prompt from each shopper who shows up. It’s building a massive data asset for whoever is running the front door to the new mall.

Finally, free traffic is what a front door monetizes well before it can charge. A 2005 study of roughly 2,500 stores in 35 U.S. malls found that 73% of anchor tenants paid no rent at all, while nearly every non-anchor lease included rent tied to sales. The anchor didn’t pay because the anchor made the traffic. Everybody else paid for a shot at what that traffic might spend on the walk between the anchors.

That’s how platform business models work. One side is always subsidized because the other side really, really needs them to drive revenue. The free referral merchants are enjoying right now may look like generosity, but it’s not. It’s the customer-acquisition phase of a business model that has run this exact play before and is building critical mass on the backs of a shopper who keeps showing up.

Why AI Shopping Agents Still Can’t Check Shoppers Out

A year ago, we all waited breathlessly for transactions to start flowing on chatbots. We’re still waiting.

The easy explanation is that the shopper doesn’t trust it. Yes and no. PYMNTS Intelligence found that 56% of consumers would let an agent search and compare products, but only 37% would let it authorize a payment to finalize the purchase. Even among the Prompt Economy Pros, the roughly 30 million people who have gone furthest in replacing search with prompts, only 23% would trust a chatbot to make the purchase decision without them in the loop.

But trust is the smaller half of the story, and it’s the half that ends up fixing itself over time just like it did with eCommerce. The bigger half is that installing the cash register is genuinely hard.

Start with the plumbing. Checkout was built end-to-end on the assumption that a human is on the other side of the screen hitting a buy button. An agent that wants to buy on the shopper’s behalf has to talk to a stack that was never designed to be talked to by software carrying a wallet. There are protocols being written to fix that. Actually, too many of them. More every month, it seems, and no agreement yet on which ones win. Every merchant watching protocols popping up has a rational reason to wait and see which ones matter. For the moment, the cost of waiting is very low.

Then there’s the part that has nothing to do with technology. The business model. When an agent buys something, who actually made the sale? That answer decides who eats the fraud, who handles the chargeback, who remits the tax, who honors the return and who gets to say they own the customer. Getting to a clean answer means integrating deep into merchant and acquirer systems, and the whole point of that integration is to leave the existing workflow intact. Order management, settlement, reconciliation, loyalty, the returns policy the call center already knows how to administer. Nobody is rebuilding any of that for agents. The task is making an agent checkout look, to the consumer and every system downstream of it, exactly like a normal one.

Retailers say that the business case isn’t there yet because the traffic isn’t there yet. No merchant builds for volume that doesn’t exist, and the volume doesn’t exist until somebody builds and consumers start showing up. They’re convinced that the chicken and egg scenario is a perfectly good reason to wait. Right up until somebody breaks it, and then it becomes a very bad reason to have waited. The 10X force Andy Grove described. And whoever breaks it first gets a head start on becoming the one front door everybody else has to pay to stand near.

But consumers are showing up, and they’ve demonstrated that checking out in the same place they’ve discovered the thing to buy is pretty good. It’s why Amazon is the world’s biggest digital retailer. The measuring stick that matters in an agentic era was never just selection. Catalogs had selection seventy years ago. It’s that the finding and the buying happen in one place, with one credential, one order history, logistics that are predictable and free, one returns policy and one company that owns the problem when it goes wrong. Absent that, it’s just a smarter way to search.

Amazon is closest because it’s the anchor and the mall. Alexa for Shopping now reaches beyond Amazon’s own catalog and Buy For Me will transact on a merchant’s site using Amazon’s credentials.

It’s the front door and the register in the same place. Google is building toward it from search. The LLMs are building toward it from the prompt. Retail, travel and health platforms are building specialty malls around occasions they already own with embedded agents that do the work inside of those closed ecosystems. Protocols are being written so agents and merchants can talk to each other about price, inventory and payment without a human in the middle.

Not all of them will succeed, and this will be a winner-takes-most outcome. There were thousands of physical malls because there’s a limit to how far anyone will drive. Agents don’t have that limit. The one that knows the shopper’s size, her budget, her taste, her favorite hotel and airline, her favorite ways to pay and how much license she’ll allow agents to take on that shopping journey follows her from one occasion to the next. A small number of front doors will own most of the starting points, and the starting point is where the shopping preferences become the habit.

3 Things Retailers Should Do Now to Prepare for Agentic Commerce

The tricky part of Grove’s 6-part framework is that the best moment to move is the moment the numbers still say you don’t have to. Once the inflection is seen in the results, the consequences become more challenging to recover from. That’s what happened at Intel. Their forecast wasn’t crazy. But it was built on the world that existed and a belief that no one could build a cheap internet access substitute. Until it showed up looking nothing like the category competitor.

Retail’s version of that forecast is already easy to see. In the Global Digital Shopping Index, only 37% of merchants said they plan to add or improve an AI shopping assistant within three years. Meanwhile, 64% of consumers said they expect to shop with agents at least occasionally within two years, and 30% expect to do it frequently or almost always. That gap isn’t one of technology’s making. It’s a lack of seeing around the corner to envision what reality will look like very soon.

What retailers can’t do is decide by not deciding, something the data suggests 63% of merchants are currently doing.

Why The Agent Risk Is the Google Search Risk

There’s one more thing, and it falls into the bucket of whether past is prologue. Google search arrived as the best complementor retail had ever seen. Free, enormous, and delivering shoppers who’d already done the deciding. Merchants optimized for it, celebrated the traffic and never once thought of themselves as paying for any of it.

The currency was the price of giving up shopper intent. Every query, every refinement, every click and every abandoned session told Google what shoppers wanted, what they compared it against and what finally closed. The merchant saw a visit. Google saw the demand curve for every category, in real time, everywhere. That’s what turned keywords into an intermediary sitting between a store and its own customer.

Over time, the free part wasn’t free anymore. Once Google owned the starting point, prominence became something merchants had to buy. Organic results moved down the page and paid ones moved up. Retail search advertising turned into a permanent line item. A price to pay to reach their own customers, priced by an auction the merchant doesn’t control and can’t audit. And it arrived one page-layout change at a time.

The agent runs the same play with better data and a shorter path. A query tells a merchant what somebody typed. A prompt tells the agent the occasion, the budget, the size, the recipient, the deadline, what got rejected and why, and which delivery promise closed the sale. Where a search page at least listed ten blue links, an agent comes back with three names. Sometimes one. There’s no page two to fall to. Prominence inside a set that small will be worth an enormous amount of money. Which is exactly why it will eventually cost retailers something.

Whether merchants end up paying a toll at the front door will also have very little to do with how good the agents get. It has everything to do with how few of them there are and how often shoppers show up. One dominant agent is the search outcome, and every merchant already knows how that story reads. Several agents competing for the same shopper, with real switching between them, is a market where a merchant still has something to negotiate with.

That points at two opportunities that are on almost nobody’s roadmap right now. The first is to keep the market from concentrating. Don’t let any single agent become the only place a catalog, real-time inventory and checkout are genuinely available. Favor open protocols over one company’s proprietary integration, plural distribution over an exclusive, and read every preferred-partner offer as the first installment on a toll. Every merchant has a perfectly rational reason to take the best deal on the table today from whichever agent has the most traffic. All of them doing that, one sensible decision at a time, is exactly how a dominant front door gets built.

The second is to back the agents whose economics point in the same direction as the merchant’s. An agent that monetizes placement is monetizing the merchant. An agent paid by the shopper, or built by a retailer, a brand coalition, an issuer, a network or a marketplace already accountable to merchant economics, is making its money somewhere else. Merchants have more say in which model wins than they think, because at this particular moment the agents need the catalogs, the inventory and the checkout far more than the merchants need the agents. That’s real leverage. But it also comes with a shelf life.

The Bottom Line: What Retailers Need to Know About AI Shopping Agents

So, let’s end by going back to my two malls. One is closing 7,900 stores this year. The other opens a new front door for every occasion a shopper can name, has no closing hours and an aisle that never ends. And it can’t yet take her money.

That last sentence is doing a lot of heavy lifting in a lot of retail strategy decks right now.

Grove’s actual lesson was never that paranoid people can’t see the future. It’s that the force that gets you is the one that shows up looking like a friend. Right now, the agent looks like the best friend a merchant has. Free, qualified, pre-sold traffic converting 54% better than anything else you can buy.

The funny part is that Intel didn’t lose mobile in a negotiation. It lost it in a spreadsheet because the thing being measured was too small to matter until it got too big to overcome.

The question for retailers right now isn’t whether agents become the new anchor store and the mall all rolled into one. It’s whether they’ll see it while there’s still something to be done about it.

The challenge is that the number telling them everything is A-OK is the same number that will keep going up right until it gets too big to overcome.

Otellini had a phrase for how that ends. “I couldn’t see it.”

Until NEXT time.

Join the 21,000 subscribers who’ve already said yes to what’s NEXT.

Why AI Shopping Agents Are Retail’s New Mall Anchor

Tags: