The End of the Retail Blind Spot: How Computer Vision and Physical AI Are Mapping the Store Aisle

NEW YORK — In the brightly lit aisle of a New York City supermarket, a man stands casually before a wall of beverages, smartphone in hand. He systematically captures a grid of photographs—shelf after shelf, label after label, facing after facing.

When asked what he is doing, he does not mention manual inventory sheets, clipboards, or estimations. Instead, he explains that he is a merchandiser for a local distributor, and this is simply "what they do now." They use artificial intelligence to scan the shelves.

The logo emblazoned on his notebook reads Manhattan Beer, one of the largest beverage distributors in the United States. The software powering his smartphone appears to be from AI Merch, a lean startup that proudly lists Manhattan Beer among its clientele. In practice, the process is deceptively simple: the representative snaps a photograph of the retail shelf, and an advanced computer vision model instantly ingests the image. Within seconds, the AI registers every sku present, identifies out-of-stock gaps, tracks pricing compliance, and audits facings.

For decades, logistics companies mapped roads, highways, and global shipping lanes. Today, through the quiet proliferation of edge AI, they are mapping the retail store.

Yet, AI Merch is merely a symptom of a much larger, highly capitalized commercial shift. To understand this phenomenon, one must look past the individual merchandiser in the aisle and examine the massive corporate machinery financing this transformation—and the staggering blind spot it aims to eradicate.

The Ultimate Retail Blind Spot

In modern retail supply chains, a profound operational paradox has persisted for generations: Brands possess precise data on what they manufacture and ship. Retailers maintain meticulous point-of-sale data on what ultimately crosses the cash register and sells. But nobody truly knows what is sitting on the physical shelf at any given moment.

The classic academic benchmark for this phenomenon remains a landmark 2002 study conducted by the Grocery Manufacturers of America, spearheaded by researchers Thomas Gruen, Daniel Corsten, and Sundar Bharadwaj. Synthesizing 52 separate studies involving more than 71,000 shoppers, the research uncovered an average global out-of-stock rate of 8.3%. When faced with an empty shelf slot, consumer behavior fractured predictably: 31% of shoppers walked out to buy the product elsewhere, 26% switched to a competing brand, and 9% abandoned the purchase category altogether.

Two decades later, the macroeconomic toll of this friction has reached astronomical proportions. According to contemporary data from the IHL Group, global inventory distortion accounts for a staggering $1.77 trillion in losses annually—with empty shelves alone representing $690.9 billion of that total. For defensive consumer packaged goods (CPG) giants like Coca-Cola, Procter & Gamble, and AB InBev, these invisible stockouts translate into continuous, silent revenue leakage.

As one fictionalized store manager quoted in industry lore puts it, capturing the traditional reporting chain: "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."

Today, computer vision is eliminating the need for human descriptions entirely.

Chronology of an Industry: From Silicon Valley Hype to Private Equity Consolidation

The evolution of retail shelf-scanning technology has matured rapidly over the past decade and a half, transitioning from expensive, experimental hardware projects to scalable software-as-a-service (SaaS) and edge-AI applications.

2010–2020: The Foundation and Venture Capital Boom

• 2010: Trax Image Recognition is founded in Singapore, pioneering the concept of digitizing retail store execution through image processing.
• Mid-2010s: Early computer vision models struggle with processing power, latency, and image clarity, limiting adoption primarily to tier-one global conglomerates running expensive pilot programs.
• April 2021: Trax secures a massive $650 million funding round led by the SoftBank Vision Fund and BlackRock, bringing its total historical venture capital funding to approximately $875 million. The market assumes an imminent, high-profile public offering.

2021–2025: Software Maturation and Strategic Partnerships

• January 2025: Enterprise technology firm Zebra Technologies integrates advanced AI agents specifically designed for automated shelf merchandising, naming retail giants like Lowe’s, Office Depot, and Total Wine as core enterprise customers.
• September 2025: Field-service management platform Repsly announces a landmark strategic partnership with ParallelDots, embedding instant image recognition capabilities directly into its workflow application. Beverage titan AB InBev and Kraft Heinz are quickly attached to the roster.
• Late 2025: Autonomous inventory-scanning robotics companies like Simbe scale operations past 3,000 active robots under commercial contract, partnering with regional grocery heavyweights like Wakefern, Schnucks, and BJ’s Wholesale Club.

2026 and Beyond: Market Correction and Private Equity Acquisition

• February 2026: In a stunning sign of market maturation, Trax sells its core image-recognition unit to private equity firm Gemspring Capital. The unit is subsequently merged with FORM, the enterprise software maker behind GoSpotCheck, creating a consolidated giant serving over 750 enterprise customers across more than 80 countries.
• End of 2026 (Projected): Walmart targets the completion of digital shelf label installations across every U.S. store. While distinct from camera-based computer vision, the initiative underscores the broader corporate crusade to turn every square inch of the physical shelf into an active, real-time data source.

Harder Than It Looks: The Engineering Reality

Despite the polished marketing campaigns put forward by enterprise tech vendors, computer vision is fundamentally more difficult than industry hype suggests.

"We always looked at the foundation of humanity, how the eyes are invented," notes Valentin Saitarli, an early founding team member of AI Merch. Now serving as co-founder and CTO of EyeX—where he applies computer vision and industrial robotics to mining and defense—Saitarli offers a sobering, technical perspective on the limitations of modern AI.

"Artificial intelligence and computer vision are actually not as sophisticated as many people think," Saitarli explains.

To illustrate his point, he uses the example of a simple beverage bottle—the exact object these retail systems spend all day counting and classifying. Holding up two identical bottles at varying distances from a camera lens, he highlights the core computational challenge: "For a machine, if I’m going to just pull it like this, they look different sizes already. Our eyes, they’ve been developing for hundreds of millions of years. We are way more advanced, way more sophisticated."

Saitarli draws a direct parallel to autonomous driving technology. "Think about Tesla. It’s a great invention, but if you look at how many training parameters are in that system, there are not that many," he points out. "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?"

The underlying bottleneck, according to Saitarli, is that humanity simply "cannot simulate reality with so many variations." In retail environments, variables such as fluctuating fluorescent lighting, shadowed bottom shelves, crumpled packaging, and customer disarray create an endless cascade of edge cases that push basic convolutional neural networks to their breaking point.

Who Is Buying? The Enterprise Adoption Landscape

Regardless of technical hurdles, the world’s largest consumer brands and retailers are deploying capital into physical AI at an unprecedented scale.

• Coca-Cola & AB InBev: Both companies are cornerstone enterprise customers of Trax, utilizing its image-recognition infrastructure to audit millions of retail facings globally. Trax claims that 30 of the top 50 consumer goods companies worldwide rely on its technology.
• Kraft Heinz: A prominent enterprise client of Repsly, leveraging the platform’s newly integrated image-recognition tools (powered by ParallelDots) to optimize in-store execution and trade promotion compliance.
• Lowe’s, Office Depot, and Total Wine: These national brands adopted Zebra Technologies’ newly deployed AI agents for shelf merchandising, streamlining the labor-intensive process of inventory auditing.
• Wakefern, Schnucks, and BJ’s Wholesale Club: Long-term partners of Simbe, deploying autonomous store-scanning robots. Wakefern reported an impressive 50% improvement in on-shelf availability across more than 90 retail locations utilizing Simbe’s robotics-as-a-service model.
• Walmart: Though utilizing electronic shelf labels rather than continuous computer vision cameras across its U.S. footprint by late 2026, Walmart represents the massive industry-wide push to digitize the physical point of sale.

Simbe CEO Brad Bogolea has championed this momentum, noting: "Retail is one of the clearest proving grounds for physical AI at commercial scale, and this is just the beginning."

Saitarli views this commercial wave as the precursor to industrial automation. "Robots will be flying all over on top of the mine, scanning the facility," he predicts, seeing a direct evolutionary path from retail shelf optimization to heavy industrial monitoring.

From VC Valuations to Private Equity Consolidation

The financial architecture supporting retail computer vision has shifted dramatically. The era of loose venture capital—exemplified by Trax’s gargantuan $650 million funding round in April 2021—has yielded to a more sober, pragmatic era defined by private equity stewardship.

When Gemspring Capital acquired the Trax image-recognition division in February 2026 and merged it with FORM, the move signaled a definitive transition from speculative growth to operational consolidation. The newly merged entity boasts a formidable footprint: over 750 enterprise customers operating in more than 80 countries worldwide.

"Bringing together FORM and Trax IR creates a more powerful, integrated platform that improves visibility, strengthens compliance, and delivers meaningful ROI," stated Ali Moosani, CEO of the newly combined company, emphasizing operational efficiency over unchecked expansion.

Saitarli’s overarching market maxim provides valuable context for these corporate machinations: "Companies are valued on future expectation." As public markets and private equity firms begin to scrutinize the actual ROI of enterprise AI spending, companies that fail to move beyond pilot programs and deliver hard efficiency metrics face painful valuation corrections.

Implications: Labor, Limitations, and the Future of the Aisle

Naturally, the rise of computer vision in retail prompts immediate questions regarding labor displacement. Will automated shelf-scanning render human merchandisers obsolete?

Promotional materials from startups like AI Merch offer a reassuring narrative for the workforce: "The sales rep still sells. The manager still manages. The distributor still runs the business."

Industry veterans tend to agree, framing AI not as a complete human replacement, but as an error-reduction engine. Saitarli, whose current work takes him into copper exploration and mining tech against a backdrop of historic commodity prices, summarizes the symbiotic relationship between machine and human labor succinctly:

"A 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."

However, observers suggest maintaining a healthy degree of skepticism toward the glowing metrics published by software vendors. As broader markets debate whether corporate AI spending is entering a cooling-off period, the true test for shelf-scanning tech will not be how many stores it can photograph, but whether those scans permanently drive down the $690 billion global cost of empty shelves.

Back in the New York City supermarket aisle, the Manhattan Beer merchandiser does not look like a harbinger of a sci-fi technological revolution. He simply looks like a worker holding a smartphone and a notebook. And that, ultimately, is the entire point: the future of physical AI is not arriving via towering humanoid robots or cinematic spectacles—it is quietly slipping into the everyday workflow, one snapshot at a time.