Be first to read the latest tech news, Industry Leader's Insights, and CIO interviews of medium and large enterprises exclusively from Retail Tech Insights
The cameras retailers bought for security are finding a much bigger job
Retail Tech Insights | Tuesday, September 22, 2026
Retailers got an unusual piece of good news this summer. Shoplifting incidents fell 12.4% in 2025, according to the latest research from the National Retail Federation.
Look a little deeper, though, and the picture gets more complicated. Half of the retailers surveyed reported an increase in repeat offenders, while 40% saw more organized retail crime activity. More than a third, 37%, also reported an increase in walkout and pushout theft.
Stay ahead of the industry with exclusive feature stories on the top companies, expert insights and the latest news delivered straight to your inbox. Subscribe today.
For retailers, that makes it harder to simply put more cameras in stores and call the problem solved. Most already have cameras watching entrances, checkout areas, aisles and stockrooms. What they have not always had is a practical way to understand what those cameras see before it is too late.
AI is beginning to change that. For example, around the checkout, an AI-powered video system can flag a register that opens with no customer in front of it, a transaction that looks nothing like the usual pattern or unusual activity around high-value merchandise and bring the event to an employee's attention as it unfolds.
While theft may be the most obvious use for all those cameras, AI is enabling many additional usage scenarios.
The checkout line is telling you something
Think about what happens around a checkout area late in the afternoon. A few shoppers become a line, and before long, six people are waiting for one register. Elsewhere in the store, a popular item is moving from a shelf faster than employees can restock it. Some shoppers stop at a new display, while others walk straight past.
These are ordinary moments in a retail store, but together they say a lot about where customers are waiting, what they are buying and how they are moving through the space. A camera that can understand what it is seeing can make some of those patterns visible, while there is still time for the store to respond.
Retailers already have ways to understand these things. Managers walk the floor, employees keep an eye on shelves and merchandising teams study sales to understand what customers are responding to. But across dozens or hundreds of stores, a lot can happen between those individual checks.
AI can make more use of the video retailers already collect. Instead of someone sitting through hours of footage to find out whether shoppers stopped at a new display, for example, AI can pick up patterns in traffic, dwell times and movement over time. A merchandising team could compare two layouts and see whether moving a display actually changed how shoppers interacted with it.
The camera can still do the security job it was installed to do with AI. However, the same footage can now tell retailers much more about what is happening on the sales floor.
Nobody needs another dashboard to watch
Store employees already have plenty competing for their attention. Anyone who has dealt with older motion detection systems knows what happens when too many alerts turn out to mean nothing. Eventually, people stop paying attention.
For alerts to be useful, the system needs enough context about what normally happens in a store to recognize when something deserves attention. Someone standing near a checkout counter at noon is ordinary. Similar activity in a restricted area after closing means something very different. Even a queue of six people could be normal during a Saturday rush and a sign that another register needs to open on a quiet Tuesday morning.
For an employee already juggling customers, shelves and checkout lines, another dashboard is hardly an improvement. A useful system should filter routine activity in the background and bring someone in when there is a reason to act.
Where that information goes matters too. A growing checkout queue might need the store manager, while a point-of-sale exception belongs with loss prevention. The value comes from getting useful information to someone who can actually do something with it.
The cameras may not be the expensive part
Large retailers rarely have one neat, standardized camera system. A chain with hundreds of locations may have accumulated cameras from several manufacturers over many years, making it difficult to justify replacing everything simply to introduce AI.
That kind of replacement may not always be necessary, since modern video intelligence can work with existing IP camera infrastructure and add capabilities through software. This is the approach we take at Lumana, allowing retailers to build on cameras they already have.
Keeping the cameras solves only part of the cost problem. High-definition video produces enormous amounts of data and continuously sending footage from hundreds or thousands of cameras to the cloud can consume bandwidth and drive up computing costs.
Some of that processing can instead happen inside the store, close to the cameras. This hybrid model allows more processing to happen on-premises so that less data needs to move to the cloud. At retail scale, AI has to make financial sense across the entire camera network, not just during a successful pilot.
A good pilot should make life easier
Retailers should resist starting with a long list of everything AI might eventually do. Pick two or three problems people inside the business already care about and give each one an owner. Queue management might belong to the store manager, for example, while point-of-sale exceptions sit with loss prevention.
The pilot should also reflect the variety of the business. A crowded urban store and a quieter suburban location can have very different customer behavior, staffing patterns and definitions of a normal day. Testing across both can reveal much more about how the technology will perform across the wider business than a polished demonstration in a flagship store.
Privacy, access and retention need to be considered at the same time, particularly as AI-powered video becomes easier to search and use to understand shopper or employee behavior. Who can access that information, what should be kept and for how long may vary by use case and local requirements.
By the end of the pilot, the useful questions are practical ones. Were checkout problems resolved sooner? Did useful alerts reach the right people? Did employees spend less time searching through footage? Did the system make running the store easier rather than give employees another tool to manage?
Retailers have already spent years putting cameras throughout their stores, capturing far more than security teams could ever reasonably watch themselves. AI creates an opportunity to make more of that video useful, whether that means responding to a checkout problem, noticing an empty shelf or understanding how customers move through the store. The cameras may have been installed for security, but their next job could extend much further across the retail operation.
Orit Dolev is an experienced Product Management Executive focused on Physical AI and 0→1 innovation. She currently serves as Principal Product Manager at Lumana, leading initiatives that transform physical environments into actionable intelligence across physical security, operations, logistics, healthcare and retail.
More in News
