While Whole Foods Market has had much success with its community-based approach to asset protection, the company is incorporating a complementary new strategy. To further reduce shrink and increase ROI, the Austin, Texas-based grocer recently started using artificial intelligence technologies, which are poised to fundamentally change how the loss prevention industry operates.
When Michael Limauro, LPC, took the helm as vice president of global asset protection at Whole Foods Market in 2021, he was facing a particularly volatile retail environment, as the country was reeling from the COVID-19 pandemic, political upheaval, and widespread crime.
To help address these challenges, he partnered with local organizations and individuals to reduce crime and increase profitability. But Limauro wanted to introduce multiple layers of asset protection strategies, so to help spearhead the AI initiative, he called on Michael Canu, an asset protection data and information technology expert.
At the time, Canu was working in the field, but his role shifted during the pandemic, and he started developing asset protection AI technologies, or, more specifically, machine-learning models.
Canu stresses that AI—especially with the rise of ChatGPT—has become a somewhat misleading marketing term that people associate with dystopian movies like The Terminator. Machine-learning is a technique that enables computers to learn from data and improve their performance over time, while the concept behind AI is that machines mimic human intelligence.
“People think AI is taking information and coming to its own conclusions,”
he said. “But ChatGPT is just a machine-learning algorithm. And that’s what 99 percent of business leaders and retail consumers interact with.”
Canu’s efforts at Whole Foods contributed to Entitlement, a machine-learning program that “essentially provides ROI on our theft events,” he said.
Implemented in January 2024, the Entitlement model bolsters Whole Foods’ case management capabilities, feeding the system a rapid stream of data points that can be quickly interpreted. Compiled information includes whether a theft event was in-store or external, if there was an apprehension or prevention, whether it was an individual or group, and what was stolen and from what department.
All these different data points are combined, which determines the ROI and helps forecast the costs and gains involved with these different theft events. Known theft events are mainly tracked for case linking and case building. If there is an incident of theft prevention, all of that data feeds into Entitlement based on the type of event.
Much more sophisticated than past methods of determining scope or stop-loss, AI has helped streamline Whole Foods’ case building. “Once we close out a case or stop a shoplifter, the magic of the machine-learning algorithm comes in,” Canu said. “It calculates ROI in a multitude of different ways and projects what the total entitlement will be over X number of days and years.”
Canu said he and his team can splice up the data sets and compare factors like retail departments and merchandise types. The technology can also forecast theft events, including potential locations and at-risk products.
Even with all these innovations, Canu stresses the important role humans play in the process. “All the data is eventually pushed to an AP person,” he said. “It still requires human interaction and a closed human feedback loop inside our network.”
Moreover, Entitlement helps save Canu’s team’s time, energy, and effort. He recently updated the Entitlement app, which enables employees to log theft events rapidly and easily using a handheld device.
“This gives us a lot more data, and it’s quicker than the old process of using a back-of-house computer,” Canu said. “Instead of having to sift through 40,000 different event types, the system is doing all that for you, and it pushes out the information you need. The fact that Whole Foods Market has adopted this shows its real value.”
Retailers Continue to Weigh Their Options
Some companies, however, are still weighing their options when it comes to AI. Brian Friedman, director of asset protection and risk management at REI, said the company is taking baby steps when it comes to adopting new AI technologies.
For case management, Friedman said the company works with the digital security provider Auror, which uses AI to help detect emerging threats and adapt. REI is also working on a customer self-serve chat project using AI, but there hasn’t been any widespread adoption beyond that. He believes that is likely to change as technology and the marketplace evolve.
“We’re an experiential company from a merchandising perspective, and there are ways we’re going to use AI to interact with customers and provide online support,” he said. “It’s going to be a huge unlock for us to gain insight quickly and put things together for case management.”
Canu said there’s likely to be rapid AI adoption over the next five to ten years, and the technology will be necessary not just for loss prevention and security, but for retailers to stay competitive. But like all technological innovations, it will come with growing pains.
“I think AI is going to sit behind a little bit of everything at some point,” Canu said. “There’s no getting rid of AI. It’s only how and where you lean into it. But the different systems will be siloed, and we’re going to find ourselves in a pickle when we realize none of these things are talking to each other. That’s one of the big AI problems we’re going to have to address.”
AI Companies Rise to Meet Challenges
Amit Kumar is working on solving some of these problems. He founded Dragonfruit AI in 2019 and serves as the company’s CEO. The company offers AI-powered video data and other applications for LP solutions, like security alerts, retail analytics, foot traffic mapping, and intrusion detection.
When it comes to the current landscape of loss prevention and AI adoption, Kumar said the most common scenario is when a retailer has an existing product or solution that is made better with AI. For example, AI-powered burglar alarms help alleviate false alarms, while AI-enabled video can do a deeper analysis of specific areas, like detecting if there’s snow or other potential hazards around a business so it can be cleaned to prevent accidents and insurance claims.
“That kind of adoption is a no-brainer,” Kumar said.
Other adoption scenarios involve new capabilities that are only possible through AI. As opposed to the laborious process of watching hours of video, AI monitoring, for example, helps prevent theft at self-checkout kiosks by using cameras, sensors, and machine-learning to analyze data and detect suspicious activity.
AI technologies are also helping create full replacement loss prevention services. With exception-based reporting that feeds into case management, you can use AI to analyze thousands of transactions quickly and efficiently and add new capabilities like voice recognition.
“None of these are examples of some crazy AI technology,” Kumar said. “Those are not getting adopted right now. If you go to a retailer and say you have an AI tool for them, they’re going to show you the door. Instead, they’re looking for cheaper, faster solutions to help solve their problems.”
Kumar said he’s seen a big change in people’s attitude toward AI over the last few years.
“People almost saw AI as black magic. Fortunately, that has changed as AI has become widespread and accessible with things like ChatGPT and Google searches. We’re seeing more pragmatic applications of AI. But retail is always careful and slow with the deployment of new capabilities.”
Like Dragonfruit AI, ThinkLP is also rolling out a new AI-powered resource for loss prevention professionals. The company, a loss and safety intelligence platform, is releasing an enhanced software module in April at the Retail Industry Leaders Association Asset Protection Conference in Washington, D.C.
Tony Sheppard is vice president of risk solutions at ThinkLP. He said AI powers some of the company’s case management and incident management offerings, as well as its exception-based reporting and analytics software. “AI integration has been gradual, but it’s certainly accelerated over the past six months,” he said.
ThinkLP’s new enhancement improves the company’s complex investigations module. Currently, the module is geared toward organized retail crime cases and enables users to collect evidence, connect incidents, and aggregate theft information, creating a case report that can be sent to law enforcement. The new version has more robust crime-linking functionality, including some capabilities powered by AI.
“You’re trying to gather as much information about an incident in the shortest amount of time possible,” he said. “AI can take information from the reports, pull out key data, and match the information to other reports.”
Sheppard stressed that while the system may indicate a suspect link or match, a human still has to confirm or reject the assertion.
“We’re being careful to ensure that the new module has human-led, opt-in functionality,” he said. “AI is still a taboo subject in some circles. While retailers are excited about the possibility of AI, they also have some concerns, especially around privacy issues and evolving legalities. In retail, you’ve got to get approval and justify why you need certain data and information.”
Bobby Haskins is senior vice president, customer – North America, for Auror. The company bills itself as a crime intelligence platform that offers tools to help reduce theft by collecting and merging detailed information on variables like suspects, locations, and items stolen.
“We use data that we collect to create a network effect not only within a single organization but with everyone using the platform inside the industry,” said Haskins, who added that about 45,000 retail locations in the US use Auror. In the past year or so, AI has become an increasingly critical part of Auror’s platform, but there are still limitations.
“It’s gotten a lot smarter, and there are a lot of benefits, but it’s not a silver bullet,” he said. “It’s a tool. And if you don’t have quality intelligence and structured data and inputs, then AI is just reviewing garbage. And garbage in is garbage out.”
Haskins points to some of Auror’s unique capabilities, like how the app employs language processing, so you can use your voice to report a theft event and easily add details like physical description and location.
The platform can also use physical descriptions and image matching to create profiles, helping identify suspects and instantly sharing the information with other users.
“Now a store manager in Ohio can track if a suspect that’s in his store is the same person who stole from a location in California,” Haskins said. “AI is helping connect the dots. It makes it easier to aggregate data and identify the people causing the most loss.”
However, Haskins said that AI does not automatically make these links and connections. It presents the information so a human who’s been trained can approve or deny a possible match.
“That’s really important in this space because there’s a lot of misunderstanding that AI is going to turn into The Minority Report,” he said. “I think AI is going to have a huge impact on the loss prevention industry. But that’s with the caveat that we need strong privacy frameworks and laws around what we should or shouldn’t do with this technology, because it is very powerful.”
