Over the past two years, AI pilots have spread across the retail industry at a rapid pace. Self checkout monitoring, customer traffic analysis, shrink detection, safety and security use cases: the interest is real and the investment is significant. But a pattern keeps emerging. The pilot works, and then nothing happens. The ROI leadership expected, whether shrink reduction, labor efficiency, or improved customer experience, never materializes.
Beth Bergmann, Chief Strategy Officer at Telaid, sat down with Brooks Thompson, Director of Emerging Technology and Partnerships, to unpack why so many retail AI pilots fail to reach scale and what the retailers who are winning are doing differently.
The Real Reason Pilots Fail
The problem is rarely the technology. As Thompson put it, "The technology does work. It just doesn't work with your specific environment."
Stalled pilots typically share a few common traits. They were run in isolation within a single team such as operations, IT, or merchandising, without cross-functional involvement. The success metrics were never tied to actual company goals. And no one seriously asked whether the solution could survive contact with a real store. Different POS systems, aging camera infrastructure, bandwidth limitations, and a workforce that was never consulted during the design phase can all undermine a solution that performed perfectly in a controlled setting.
Then there is the sustainability question. Making a change is one thing. Keeping it in place long enough to see results is another. As Thompson noted, "You can make a change, but if you don't keep with it, then you don't make the ROI that you signed up for." That requires not just deployment, but ongoing measurement, monitoring, and management.
A Real-World Example: Self Checkout Shrink
Bergmann walked through a recent engagement with a large retailer that had rolled out self checkout and suspected it was driving increased shrink, but could not quantify how much, where, or when.
The root issue was not the technology. The teams responsible for operational improvement and asset protection had never been brought into the conversation during the rollout. There was no structured hypothesis and no clear ownership of the problem.
Telaid stepped in and treated it as a proper test. They assessed three locations across the retailer's national footprint, used AI models to analyze footage, and built a prediction model covering both intentional and unintentional theft across stores at different stages of self checkout adoption, ranging from six months in to eighteen months in. The finding was concrete: approximately 3.5% of shrink leaving the store was attributable to self checkout. With that number in hand, the retailer had what it needed to make a business decision rather than continuing a technology debate.
Starting Right in a Non-Greenfield World
Most retailers are not building from scratch. They are retrofitting existing stores, working with legacy systems, and balancing current technology investments against future roadmaps. That reality shapes how Telaid approaches every engagement.
Thompson described the process: start with the current technology stack and understand what is already in place, what has been purchased, and what is on the roadmap. Only after establishing that foundation does it make sense to discuss use cases, compute requirements, and the ROI models that will justify moving forward.
"A lot of people start with the AI review first, and they get excited about stuff and want to move forward," Thompson said. The risk is investing in a solution and discovering late in the process that it cannot scale or that the infrastructure cannot support it. Getting the infrastructure conversation right at the beginning avoids that outcome entirely.
Starting Right in a Non-Greenfield World
Most retailers are not building from scratch. They are retrofitting existing stores, working with legacy systems, and balancing current technology investments against future roadmaps. That reality shapes how Telaid approaches every engagement.
Thompson described the process: start with the current technology stack and understand what is already in place, what has been purchased, and what is on the roadmap. Only after establishing that foundation does it make sense to discuss use cases, compute requirements, and the ROI models that will justify moving forward.
"A lot of people start with the AI review first, and they get excited about stuff and want to move forward," Thompson said. The risk is investing in a solution and discovering late in the process that it cannot scale or that the infrastructure cannot support it. Getting the infrastructure conversation right at the beginning avoids that outcome entirely.
Where Retailers Are Winning
Checkout remains the highest-priority use case for most retailers, covering shrink validation and labor optimization at the point of sale. But the broader conversation has evolved. A few years ago, the focus was on pinpoint solutions: solve one specific problem and move on. More recently, retailers shifted toward platform consolidation, trying to handle everything with a single vendor. What Telaid is seeing now is a third phase: solutions that integrate with existing infrastructure through open APIs and shared data, rather than replacing what is already there.
Distribution centers are also emerging as a high-impact area. The scale of ROI available in a DC is significantly larger than in an individual store, with fewer facilities to manage, making the investment easier to scale. Active use cases include truck path optimization, yard visibility, and gate tracking.
Labor optimization is the use case Thompson sees rising to the top in 2026. For retailers and quick service restaurants alike, labor is the number one controllable expense. Any solution that removes tasks or helps close the gap between planned and actual labor hours delivers direct and measurable financial impact.
The Mindset That Separates Winners from Stalls
Bergmann closed the conversation with a clear takeaway. The retailers succeeding with AI are not running more pilots. They are designing for scale before they call it a pilot, ensuring the right teams are engaged from the start, and building toward a clearly defined North Star rather than chasing the latest technology.
Leadership alignment matters too, Thompson added. It is less about leadership validating every technical decision and more about everyone agreeing on the metric the organization is trying to move. When that alignment exists, the right technology gets adopted and sustained.
"Don't just trial everything," Bergmann said. "Be purposeful, intentional, understand your environments, and then bring the right partners to the table to help you execute your business."
Ready to take the next step? Contact us below to see how we can partner on your next project for implementing AI and syncing technology across your stores.
