AI in Physical Retail: What's Actually Working in 2026

There's a specific type of retail technology that achieves significant conference floor presence, generates enthusiastic trade press coverage, and then quietly disappears from most stores within 18 months. Personal shopping assistants on tablets. QR code discovery journeys. Virtual reality in changing rooms. Each of these had their NRF moment, their pilot announcement, and their subsequent disappearance into the footnotes of technology retrospectives.

I'm not saying physical retail AI is in that category. I am saying it's worth being specific about which parts of it are genuinely scaling and which parts are still looking for the retail context where they actually deliver ROI.

What's Actually Scaling

Computer vision for inventory and shelf intelligence is the physical retail AI application I'm most consistently hearing positive things about from practitioners, as opposed to vendors. The use case is unglamorous (knowing what's on the shelf and where, in real-time), but the operational value is substantial. The British Retail Consortium's 2025 Retail Crime Survey put UK customer theft losses at £2.2 billion in the year to August 2024, with over 20 million theft incidents recorded. Out-of-stocks cost sales and frustrate customers. The labour cost of manual shelf scanning is ongoing. A computer vision system that gives continuous accurate shelf state and flags anomalies in real time is addressing problems that have existed for decades.

The technology has matured enough that the industry is converging on a hybrid architecture: overhead cameras, shelf sensors, and AI analysis working together rather than any single sensor type carrying the full load. The early single-sensor approaches, whether just cameras or just weight sensors, all had failure modes. The layered approach is more robust.

Demand forecasting and planogram optimisation using AI has been in production at large grocery retailers for several years and at this point is fairly mature technology. Tesco, M&S, and Sainsbury's have all been refining AI-driven planning models across their store estates. The interesting development in 2026 is the integration with supply chain AI that handles tariff disruption and sourcing volatility, with demand and supply modelling becoming more tightly coupled.

Staff scheduling and task management using AI is producing ROI in the mid-market as well as enterprise, and it's a category that doesn't get much press because it's not photogenic. With 2025 bringing steep increases to both the National Living Wage and National Insurance Contributions, UK retailers face a combined £5 billion rise in labour costs, creating pressure to get scheduling right. Workforce tools that match staffing levels to predicted foot traffic patterns, with real-time adjustment for the actual day, reduce both under-staffing and over-staffing. The data is internally available, the model is not particularly complex, and the operational savings are measurable. This one is working.

What's More Mixed

Smart carts are the most visible physical retail AI application and also the one with the most nuanced results picture. Instacart and Morrisons announced a UK-first pilot of Caper Cart smart trolleys, launching at one Morrisons store in early 2026 with potential for wider rollout. A single-store pilot by a major UK grocer is a meaningful signal of intent. But a March 2026 piece in Retail Technology Innovation Hub reports what Sylvain Perrier of Vertus Group heard from a store associate at another retailer that had already tried and removed the carts: "It was a solution looking for a problem… we had them removed due to lack of use and inaccuracies with lightweight products."

The honest picture is that smart cart adoption rates vary by store context. High-volume urban stores where customers are comfortable with technology and doing a considered shop of reasonable size show better engagement than smaller stores or those where the customer base is less tech-comfortable. The ROI calculation also depends heavily on labour costs and shrinkage, with the business case stronger in high-wage markets with significant losses than in lower-cost contexts.

Autonomous checkout at the store level has had a complicated few years. Amazon's pullback from Just Walk Out in its Fresh stores in 2024 was one signal; the company's decision in January 2026 to close all Amazon Go and Amazon Fresh physical stores entirely is a stronger one. The technology itself has improved steadily. The business case at general grocery scale has not followed.

Where autonomous checkout appears to be finding its footing is in high-footfall, smaller-format venues: stadium concessions, transit hubs, campus stores, where staff costs per transaction are unusually high and basket sizes are predictable. Zippin operates in this space, with around 96 locations globally as of early 2026. That's a real market. It's just not the market the conference demos suggest.

The UK High Street Lens

Physical retail AI plays out differently on the UK high street than it does in US grocery or US big-box. The UK retail estate is older, stores are smaller, and the economics of fitting computer vision systems into a 1950s shop unit are different from fitting them into a purpose-built 2010s distribution centre.

The UK's high street is also in a structural squeeze. Declining footfall, rising business rates, and labour cost increases from National Living Wage rises all create pressure to adopt labour-saving technology, while simultaneously limiting the capital available to invest in it. The retailers doing this well in the UK tend to be those with strong central data infrastructure (which the major grocery chains have) or those where the physical estate is small enough and standardised enough that technology rollout is manageable.

Independent UK retailers, the kind of business that makes up a significant portion of the high street, are largely not participating in physical retail AI at any meaningful scale, and there isn't a clear path to that changing soon. The access gap that exists in digital commerce has a physical retail equivalent.

The Honest Bottom Line

Physical retail AI is not a single thing. The conference version, with sleek stores, frictionless checkout, and robots gliding down aisles, is a real vision but a partial one. In 2026, the parts that are clearly working are mostly invisible: demand forecasting, staffing optimisation, shelf sensing. The parts that are visible on the shop floor are still finding their scale.

Which is perhaps how it should be. The technology that actually transforms retail operations is rarely the technology that gets the conference booth.