Coca-Cola, AB InBev and Kraft Heinz are quietly turning store shelves into AI data

In a store at a New York City mall, a man stood in the aisle taking pictures of the shelves, one after another.

Asked what he was doing, he said he was a merchandiser for a local distributor. That's what they do now, he said. They use AI to scan the shelves.

The logo on his notebook said Manhattan Beer, one of the biggest beverage distributors in the country. The tool appears to be from a small startup called AI Merch, which lists Manhattan Beer as a customer on its website. The rep snaps the shelf and the AI reads it, what's there, what's missing, the prices, the facings. So instead of mapping the roads, they map the stores.

But AI Merch is just one example. The bigger story is who's paying for this, and why.

The shelf is the blind spot

Brands know what they shipped. Stores know what sold. Nobody knows what was actually on the shelf.

The classic benchmark is a 2002 Grocery Manufacturers of America study by Thomas Gruen, Daniel Corsten and Sundar Bharadwaj: 52 studies, 71,000+ shoppers, an average out-of-stock rate of 8.3%. Faced with an empty slot, 31% bought elsewhere, 26% switched brands, 9% bought nothing.

IHL Group's current figure: $1.77 trillion in global inventory distortion, $690.9 billion of it empty shelves. A real hit for defensive names like Coca-Cola.

The film's manager character puts the old reporting chain plainly: "By the time it gets to me, I'm no longer looking at the shelf. I'm looking at somebody's description of the shelf."

Four ways to see a shelf

• Rep phones: AI Merch, Repsly, FORM.
• Shoppers: Storesight, crowdsourced photos, 80% of all-commodity volume claimed. CEO Ty Kasperbauer: "The data needed to win at the shelf is now non-negotiable."
• Robots: Simbe's Tally, several aisle passes a day.
• Store handhelds: Zebra Technologies.

Harder than it looks

"We always looked at the foundation of humanity, how the eyes are invented," said Valentin Saitarli, who was part of AI Merch's founding team. He is now co-founder and CTO of EyeX, where he works on computer vision and robotics for mining and defense. He is blunter about the technology than most of the people selling it. "Artificial intelligence and computer vision is actually not that sophisticated as many people think," he said in an interview in July.

His example was a bottle, which is exactly what these systems spend all day counting. He held up two identical ones at different distances. "For a machine, if I'm going to just pull it like this, they look different size already," he said. "Our eyes, they've been developing for hundreds of millions of years. We are way more advanced, way more sophisticated."

"Think about Tesla. It's a great invention, but if you look how many training parameters are in that system, there are not that many," Saitarli said. "It will take a 16-year-old kid two weeks to drive a car. How many years does it take Tesla to teach the machine to drive?"

As Saitarli put it, "We cannot simulate reality with so many variations."

Who is buying

Coca-Cola and AB InBev. Both Trax customers. Trax says 30 of the top 50 consumer goods companies use it.

Kraft Heinz. A named customer of Repsly, which added image recognition in a September 2025 partnership with ParallelDots. AB InBev is on that list too.

Lowe's, Office Depot, Total Wine. Named as customers when Zebra added AI agents for shelf merchandising in January.

Wakefern, Schnucks, BJ's Wholesale. Simbe customers. Simbe says Wakefern improved on-shelf availability 50% across more than 90 stores. 3,000+ robots under contract. CEO Brad Bogolea: "Retail is one of the clearest proving grounds for physical AI at commercial scale, and this is just the beginning." Saitarli sees the same thing coming for heavier industry: "Robots will be flying all over on top of the mine, scanning the facility."

Walmart. Digital shelf labels in every U.S. store by end of 2026. Not cameras, but the same push to make the shelf a data source.

From $650 million to private equity

Trax: founded 2010, Singapore. April 2021: $650 million round led by SoftBank Vision Fund and BlackRock, about $875 million raised in total. February 2026: image-recognition unit sold to Gemspring Capital and merged with FORM, maker of GoSpotCheck. Combined: 750+ customers, 80+ countries.

"Bringing together FORM and Trax IR creates a more powerful, integrated platform that improves visibility, strengthens compliance, and delivers meaningful ROI," said Ali Moosani, CEO of the combined company.

Saitarli's general rule on valuations applies here too: "Companies are valued on future expectation."

The limits

The numbers deserve some skepticism, since most of them come from the vendors. And markets are already arguing over whether AI spending slows.

The labor question is the obvious one. AI Merch's own promo film answers it this way: "The sales rep still sells. The manager still manages. The distributor still runs the business."

Saitarli, who's now taking the same tech into copper exploration with copper near record prices, put it this way: "Machine will prevent lots of mistakes at first. And then, when a machine cannot make a decision, a human will explain to the machine what that parameter is actually about."

What to watch

• Sells the tools: Zebra.
• Buys the data: Coca-Cola, AB InBev, Kraft Heinz.
• Runs the stores: Lowe's, BJ's, Walmart.
• Private: Trax/FORM, Repsly, Storesight, Simbe, AI Merch.

The merchandiser in the aisle didn't look like the future of AI. He looked like a guy with a phone and a notebook. That's the point.