Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Organized retail crime (ORC), fraud and theft continue to plague retailers, despite the development of loss prevention teams, time spent locking up goods and other strategies meant to deter crimesters.
But technology may become a more popular answer to those threats. California-based Appriss Retail, which was just acquired by Gemspring Capital, has developed artificial intelligence-based technology to mitigate crime and fraud, both digitally and in store.
Pedro Ramos, the company’s chief revenue officer, spoke with Sourcing Journal about the current trends in the retail crime landscape, what AI’s place is in aiding retailers in preventing and reacting to bad actors and more; here’s a look into the conversation.
Sourcing Journal: Could you talk through some of the trends you started to see in retail theft and organized retail crime over the past couple of months, whether that’s inclusive of 2024 or exclusive to 2025?
Pedro Ramos: What we see in our application, our company, is there is an evolution of return and claim fraud—from the occasional [individual] or abusive entity to professionals entering this space. We see countless examples of systemic return fraud…and all kinds of [examples, like] tender manipulation, fraudulent receipt use and all kinds of other behaviors that are highly indicative of somebody who’s well organized and knows what they’re doing.
We also see large groups or entities that appear to be the same person, but if you look at their total number of accounts generated, number of attempted uses of multiple credit cards and IDs, they’re clearly the same entity. Our AI engines are able to identify those, so we see that growing. It’s easy—it’s a pretty low-risk crime, and it converts into cash. We see a lot of return redemptions to multiple debit cards, and I’m just going to make the assumption that those debit cards, those accounts are being closed, and the accounts are being closed.
From conversations with retailers across the board, ORC continues to be an issue or a topic. What they are seeing is a little bit of a change. The thing that blew up over the past few years—smash-and-grabs and so forth—seems to have declined, but ORC is [still] in peak and not declining.
SJ: We’ve established a little bit so far about the problem that retailers are having, both with digital fraud and with organized retail crime in stores. I’m curious to get your perspective on how technology is being deployed today, both online and in stores to either prevent or deal with proactively retail crime before it ever happens?
P.R.: There are sophisticated technologies in store that are being deployed now. There’s [also] been a lot of public media around product protection, which is basically securing high-risk products. It depends on the individual retailer’s strategies and philosophies, but that seems to be a growing trend across the industry, and it’s going to impact the consumer. It’s also going to, inevitably, impact sales…
It’s [all about] staying ahead of these bad actors. They’re anonymous, so there’s very low risk for them. There’s plenty of training material on the internet to teach newcomers on how to become professional organized retail criminals.
The solutions that are effective in [preventing crime] are real-time AI solutions that are able to identify these entities accurately. You want to do no harm to good customers, and then effectively shutting down those [illegitimate] transactions.
SJ: You started to speak there about something I think is really interesting, and a trend we’re starting to hear about, which is being able to separate bad actors from regular or loyal customers. Could you share a bit about how, specifically, AI and technology are able to help retailers make that distinction?
P.R.: Today, the consumer interacts with a brand through every single channel that brand has, and it’s generally online, in store and through a call center. All those channels need to be making decisions off of the total consumer profile—so a 360 view of the consumer. That’s the first thing.
The challenge is, today, many retailers still try to use policies to address this issue—or rules engines or technologies that are ‘digital protection,’ but they’re not. They’re just policies that are implemented via a digital mechanism. The challenge there is they’re usually focused on either volume or velocity—how much returning [is done] or claims are submitted within a window, or how frequently are these being done by the consumer identity? Without taking the consumer’s total profile, it leads to very high insult rates, and so you’re hurting your consumer…
AI is able to look at the consumer across all channels across the landscape and say, although they may be [making more] returns, their total purchases are much higher [and] they’re a high-value consumer—go ahead and approve the transaction. On the completely opposite side of that spectrum, AI is able to say, I’ve never seen this consumer. This is not a consumer, in fact—this entity has got a negative total sales, which means they’ve returned more product than they’ve ever purchased, and we’re going to decline the transactions.
AI is well suited when applied to solving very specific problems, and then it actually becomes exponentially better if it’s a supervised model, like ours, where we’re constantly feeding input back into the AI models based on real-life examples.
SJ: You mentioned how total transaction value, positive or negative, can help an AI model to understand whether the consumer is a regular consumer or a bad actor, a bad entity. Could you say a bit more about the type of data that helps an AI model to flag whether a purchase should be approved or denied?
P.R.: At a very high level, you’re looking at total purchase data and total negative transaction data. That means voids, returns, promotions, how many debit accounts, credit card accounts, the loyalty data and then ultimately some type of ID mechanism for folks who do non-receipt returns, whether it’s a simple phone number used in order to identify that person [or otherwise].
The other really cool thing about these applications is that, the retail world, and the world outside of retail’s four walls, doesn’t stand still. So as new challenges emerge, once these applications are in place, we personally see our retailers coming to us saying, ‘We’re seeing new trends,’ or we’re going to our retailers and saying, ‘We’re seeing a new pattern of behavior. Would you like to solve that pattern?’ Generally, the data is there. It’s the algorithms that are starting to recognize that pattern, and then we can solve those problems over and over again.
SJ: We’ve chatted a lot so far about preventative measures for organized retail crime and digital theft, but I’m also curious to get your take on measures taken after a crime already occurred. Is AI as robust of a system to use for that kind of situation, or are there other technologies or barriers you would cite as more effective for post-retail crime scenarios?
P.R.: Generative AI is really well suited [for this]. We use it in our ORC module for case management…and generative AI, along with other AI models or algorithms, [can] quickly consolidate evidence, whether it’s written, whether it’s structured data in a database, whether it’s images and so forth. We’re able to quickly scour them and say, ‘Okay, I have X number of events that look like they’re the exact same person, exact same product line.’
That makes end users exponentially more effective and allows for the detection of people that would otherwise be treated as low-level shoplifters, and allows for those incidents to be tied together into larger organized retail crime cases that can be brought up to higher courts or law enforcement bodies that are willing to take them on. You can basically take out [ORC] rings at scale.
Feeding some of that imagery back to camera systems that can recognize known offenders and alert store management…is also a preventative measure.
There is one measure that nobody ever talks about which is really, really effective: inventory controls.
SJ: Say a little bit more about inventory control. If nobody else is talking about it, why don’t we?
P.R.: I was a retailer for 25 years. A good way to prevent the ORC is to make it really difficult for them to walk away with a large grab. Having really good—and AI helps here, as well—inventory management systems that identify…targeted items in order to reduce the exposure of those items on the shelf without impacting sales is a very effective tool.
It is going to require a little bit more labor to restock those shelves, but if you’re a professional organized retail criminal, you’re exposing yourself every time you commit a shoplifting act, so you want the payback to be as significant as possible. If somebody’s effectively managing the exposed inventory of these high-risk items, it’s just going to make it a less attractive target…
It’s going to require…really effective inventory management systems and shelf allocation, all of which the modern [retailers] are starting to use AI to do. [If I’m a crimester,] I can walk in and I have a minimal amount of product, just enough to keep the display looking perfect enough to sell, so my payback for the risk I’m about to take is significantly lower. I’m always going to target the company with the largest exposure.
