Computer Vision Promised Retail a New Pair of Eyes. The Real Problem Was Always the Brain.

Kroger has deployed visual AI at 1,700 grocery stores and counting. The system watches self-checkout lanes with high-resolution cameras, integrates structured POS data with unstructured video feeds, and flags scanning errors in real time. According to Chain Store Age, more than 75 percent of self-checkout errors at those locations are now corrected without employee intervention. That is a real operational outcome, documented across a real fleet of stores, solving a real problem that costs the grocery industry billions of dollars a year.

And yet.

The harder question is not whether computer vision can see what happens inside a store. By now, that much is settled. Research and Markets values the computer vision for retail market at $5.24 billion in 2026, growing at roughly 24 percent annually. Gartner projects worldwide AI spending will reach $2.52 trillion this year, and the firm’s analysts have described the current phase of enterprise AI adoption as one where proven outcomes matter more than speculative potential. The technology is maturing. The installed base of cameras, edge servers, and inference models is expanding across every major retail format. None of that resolves the central tension in physical retail: generating data has never been the problem. Acting on it before the moment passes has always been the harder part.

For CPG leaders, retail media strategists, and operations executives managing portfolios across multiple retailers, this tension carries specific and underexplored implications that the current wave of computer vision coverage largely ignores.

When the Retailer Sees the Shelf and the Brand Does Not

Shelf monitoring is computer vision’s most commercially mature retail application. Walmart Canada rolled out camera-based inventory detection across its entire Canadian store fleet after a 70-store pilot. According to Vision Systems Design, the system uses computer vision models trained on more than 1.4 billion labeled images to identify out-of-stock conditions, misplaced products, and planogram deviations. Asda in the U.K. has trialed shelf-mounted cameras that each monitor roughly eight feet of product space, as reported by RetailWire. Retailers from Woolworths in Australia to Starbucks in North America are building similar capabilities, combining cameras, edge AI, and integration with existing replenishment systems.

The dollars behind the problem explain the investment. IHL Group’s September 2025 research found that global inventory distortion, the combined cost of out-of-stocks and overstocks, still accounts for $1.73 trillion in annual losses across retail. Even modest improvements in on-shelf availability produce measurable revenue gains, because the consequences of an empty shelf compound quickly: a shopper who cannot find what they came for may switch categories, switch brands, or switch retailers entirely.

The problem for CPG brands is straightforward. The retailer controls the cameras, the data pipeline, and the decision about what happens with the resulting intelligence. A computer vision system that detects an empty shelf at 2 p.m. on a Tuesday is valuable only if that signal reaches someone who can act on it, whether that means pulling product from the backroom, adjusting an order, or flagging a distribution issue. If the brand whose product is missing from that shelf learns about the gap three days later through a syndicated data report, the sale is already lost.

Retailers investing in computer vision are building a real-time view of store conditions that their supplier partners typically cannot access at the same speed or granularity. For brand teams responsible for retail execution across hundreds or thousands of stores, the question is whether computer vision data becomes a shared resource that improves joint business planning, or a proprietary asset that widens the information gap between retailer and supplier.

Category captains and large CPG companies with strong joint business planning relationships may eventually negotiate access to shelf-level visibility data. Smaller brands and mid-tier suppliers are less likely to receive that access unless they can demonstrate a clear value exchange. The companies that recognize this dynamic early, and begin structuring their retailer conversations around data-sharing frameworks rather than simply hoping the technology trickles down, will be better positioned as computer vision deployments scale.

Shrink Technology Outpaces Shrink Strategy

Loss prevention is the second pillar of retail computer vision, and the investment momentum is substantial. The NRF’s 2025 Impact of Theft and Violence report, based on surveys of retail companies representing $1.3 trillion in annual sales, found that shoplifting incidents increased 18 percent from 2023 to 2024 and that violence tied to theft events rose 17 percent over the same period. Trigo Vision, an Israel-based computer vision firm, launched a loss prevention solution in mid-2025 designed to cross-reference items shoppers pick up with items scanned at checkout and send real-time alerts when discrepancies surface.

Kroger’s deployment of visual AI at self-checkout is explicitly framed as both a customer experience improvement and a loss prevention tool. Chris McCarrick, Kroger’s senior manager of asset protection solutions and technology, described the system to Chain Store Age as one that strengthens checkout accuracy without creating friction for shoppers. The company reports reduced shrink alongside smoother self-checkout operations.

But several tensions remain unresolved. First, accuracy. Industry reporting has documented persistent false-positive problems across computer vision loss prevention systems. Alerts triggered by ordinary shopping behavior, like picking up multiple items in quick succession, or by produce that scans inconsistently due to weight variations, remain common complaints from retailers that have deployed these tools. False positives consume associate time, create friction for honest shoppers, and erode the efficiency gains the technology is supposed to deliver.

Second, the regulatory picture is shifting underneath these deployments. Twenty U.S. states now have comprehensive consumer privacy laws either in effect or enacted, according to Gibson Dunn’s 2025 U.S. Cybersecurity and Data Privacy Review, with biometric data receiving heightened scrutiny in many of them. The FTC has also scrutinized in-store facial recognition. Computer vision systems can be designed to track shopper behavior using non-identifying details rather than biometric identification, but the regulatory environment is moving quickly, and retailers deploying these systems need governance frameworks that can adapt as rules change.

Brand teams involved in shrink conversations should recognize how they intersect with retail execution in ways that are not always obvious. When retailers respond to loss by locking products behind cases, adding friction to self-checkout, or reducing store hours, brand sales velocity suffers. Computer vision offers a potential alternative: intervening earlier, more selectively, and with less disruption to the shopping experience. But that outcome depends on how the technology is deployed, not simply on whether it exists. Brands with a seat at the table in loss prevention discussions, particularly in categories with high shrink rates, should be pushing for implementations that protect the shelf while preserving shopability.

The Checkout-Free Mirage and What It Taught the Industry

Amazon’s experience with Just Walk Out technology offers a useful corrective to the assumption that computer vision in retail follows a linear adoption curve. In January 2026, Amazon announced it would close all 72 of its Amazon Go and Amazon Fresh stores, as reported by Grocery Dive and GeekWire, shifting focus to Whole Foods and online grocery delivery. The company had already removed Just Walk Out from its U.S. Amazon Fresh stores in 2024 after finding that customers preferred checkout options that let them see their spending in real time, replacing the system with smart shopping carts.

The technology worked. It simply did not fit the format. Just Walk Out found genuine traction in smaller, high-throughput environments. Amazon reported in an April 2024 blog post that Lumen Field, home of the Seattle Seahawks, saw an 85 percent increase in transactions and a 112 percent increase in per-game sales after deploying its first checkout-free concession store. By late 2024, more than 140 third-party locations in four countries were using the technology, concentrated in stadiums, airports, hospitals, and college campuses.

The lesson is not that cashierless checkout failed. What matters is that computer vision applications must match their format. A technology that excels in a 2,000-square-foot concession stand with mission-driven shoppers buying a handful of items may not translate to a 50,000-square-foot grocery store where shoppers build baskets over 30 minutes and want to track their running total. Retailers evaluating computer vision investments should resist the temptation to extrapolate from headline use cases and instead assess format fit rigorously, store by store and function by function.

This matters to brands because the checkout format shapes shopper behavior in ways that directly affect basket size, impulse purchasing, and promotional redemption. A frictionless checkout in a grab-and-go format favors speed over discovery. A smart cart in a full-size grocery store that displays a running total and suggests related items creates a different set of commercial opportunities entirely. Understanding which computer vision implementations are operating at which retailers, and how those implementations change the in-store experience, is now a necessary input for trade marketing and channel strategy.

The Data That Never Leaves the Store

The most significant long-term implication of computer vision in retail is not any individual application. It is that physical stores are, for the first time, generating a continuous data stream comparable to what e-commerce platforms have produced for over a decade.

Until recently, physical retail operated with periodic snapshots: syndicated data arriving weekly, manual audits conducted monthly, planogram compliance checks performed during field visits. Computer vision replaces that cadence with continuous measurement. Foot traffic patterns, dwell times, shelf conditions, queue lengths, and conversion rates can all be captured in real time and fed into operational workflows.

The question for CPG leaders is whether this data will be shared, sold, or kept behind the retailer’s firewall. Retail media networks have already demonstrated that retailers can monetize first-party data from their loyalty programs and e-commerce platforms. Shelf-level computer vision data, which captures how shoppers actually interact with products in the physical store, is potentially even more valuable. A retailer that can tell a brand exactly how many shoppers paused in front of its display, picked up a product, and then put it back is holding a data asset that no syndicated provider currently offers at that level of granularity.

Some retailers are already moving in this direction. Shelf intelligence platforms designed to monetize real-time shelf data for CPG partners are entering the market, suggesting that computer vision data may eventually flow through the same commercial channels as retail media impressions, priced and packaged for brand partners willing to pay for in-store visibility.

For brands accustomed to negotiating trade spend, promotional calendars, and retail media budgets, this represents a new category of investment. The organizations that build the analytical capability to connect in-store behavior data with their existing shopper insights, trade promotion analysis, and media measurement frameworks will extract disproportionate value from these partnerships. Those that treat shelf-level data as an IT procurement decision rather than a commercial strategy question will pay for information they do not know how to use.

Building the Evaluation Framework Now

Gartner’s John-David Lovelock said in January 2026 that AI will most often reach enterprises through their existing software providers rather than through speculative new projects. That framing applies directly to computer vision in retail. The technology is past the proof-of-concept stage. The relevant decisions now center on integration, governance, and commercial terms.

Retailers evaluating or expanding computer vision deployments should be asking three questions that the current vendor conversation rarely surfaces. First, how does this system integrate with existing replenishment and labor scheduling workflows? A camera that detects an empty shelf but routes the alert to a dashboard no one checks during peak hours has produced data, not action. Second, what governance framework applies to the behavioral data this system collects, and does it account for the accelerating pace of state privacy legislation? Third, if this data has commercial value to supplier partners, who owns it and under what terms?

On the supplier side, the parallel set of questions is equally specific. Which of your top retail partners are deploying computer vision, and do your joint business planning agreements include provisions for accessing the resulting data? Are your field teams and category managers equipped to interpret and act on real-time shelf intelligence, or will this information arrive faster than your organization can process it? Does your trade investment framework account for the possibility that in-store visibility data will become a paid tier within retail media networks?

The technology that lets retailers see their stores in real time is arriving faster than most organizations have built the operational capacity to respond to what it reveals. The companies that will benefit most are not necessarily the ones with the most cameras installed. They are the ones that have figured out what to do in the fifteen minutes between when a camera spots a problem and when a shopper walks away empty-handed.

Conversations On Retail

Conversations On Retail is a gathering place and resource center for retail and CPG executives, built to make it easier to stay current, discover the technologies and solutions shaping the industry, and connect with the people driving it forward.

We publish news, views, and reviews from staff editors, contributing experts, and trusted partners. Some articles are developed internally, while others are submitted by industry contributors or adapted from interviews and recorded conversations with industry leaders.