Latest / Internet of Things with Fexingo: Connected Devices, Sensors, and Industrial IoT / How IoT Sensors Are Preventing Retail Theft Without Facial Recognition
Transcript
- Lucas: If you've walked into a grocery store in the past couple of years and noticed that the self-checkout area feels a little more... supervised — you're not imagining it. But the technology behind that isn't what most people assume. Luna: Yeah, a lot of us default to 'cameras and AI facial recognition' as the go-to solution. But you're saying there's something more... physical? Lucas: Exactly. There's a whole category of IoT-based loss prevention that uses pressure sensors, RFID tags, and smart shelves — no cameras, no biometric data. And one regional grocery chain, Schnucks out of St. Louis, has been quietly testing this approach in 14 of their stores. Luna: Fourteen stores — that's a decent pilot. What kind of results did they see? Lucas: They reported a 22 percent reduction in theft in the first six months. But here's the interesting part — the system they're using, from a company called Ceptology, doesn't try to identify shoplifters. It focuses on the items. Luna: So walk me through how that actually works for someone with a cart full of groceries. Lucas: Sure. Picture a standard self-checkout lane. You scan your items, you bag them, you pay. Ceptology installs a set of load sensors under the bagging area — essentially, a very precise scale. And they also put RFID readers near the exit and pressure mats on the floor. Luna: So the system knows the weight of every item in their database? Lucas: Exactly. When you scan a can of beans, the system expects a certain weight to appear on the bagging scale within a fraction of a second. If the weight doesn't match — either because you didn't scan something and put it in the bag, or you scanned a cheap item and bagged an expensive one — the system flags it. Luna: And then what happens? Lucas: That's the clever part. The system doesn't trigger an alarm or call security. It just sends a subtle alert to the store associate's handheld device — something like 'Lane 4, weight discrepancy on item category: deli meats.' The associate can walk over, look at the screen, and say 'Hey, did that package of prosciutto scan okay?' Luna: It's de-escalation by design. They're not accusing anyone — they're inviting a correction. Lucas: Exactly. And that's key because one of the biggest headaches in retail loss prevention is false accusations. If a security guard wrongly accuses a customer of theft, you've got a lawsuit. But if the system just flags a weight anomaly and the associate politely asks if everything scanned properly — it's a service interaction, not an accusation. Luna: Did Schnucks see any change in false-accusation incidents? Lucas: They said the number of escalated incidents — where a customer complained about being treated like a thief — dropped by 40 percent in the pilot stores. Because the interaction is framed as 'let me help you make sure you're not overpaying' rather than 'I think you stole something.' Luna: That's interesting — they actually flipped the script. Instead of assuming guilt, they're assuming a scanning error. Lucas: Right. And the sensors work as a deterrent too. Ceptology found that once customers see the 'bagging area is monitored by weight sensors' sign, the intent to steal rate drops significantly. The system doesn't need to catch anyone — it just needs to make theft feel risky. Luna: And you mentioned RFID at the exit — what's that for? Lucas: That's a backup layer. Some high-value items — like razor blades, baby formula, electronics — have RFID tags. When a customer walks through the exit gate, the readers check whether that tagged item was legitimately purchased. If it wasn't, the system logs it and can alert staff. But again, no facial recognition, no video storage. Luna: So the whole system is built around physical measurement — weight, proximity, radio signals — not visual identification. Lucas: Exactly. And that has real implications for privacy advocates. There's no database of shopper images, no biometric profile. The sensors are essentially blind to who you are — they only care about the relationship between what's scanned and what's in your bag. Luna: I can see that being a selling point for certain demographics. I know a lot of people who'll avoid stores that use facial recognition. Lucas: And that's exactly the niche Ceptology is targeting. Their pitch to retailers is: 'You get the loss prevention improvement without the PR risk of being the store that scans people's faces.' It's a smart positioning. Luna: Let's talk about the cost. What does something like this run a retailer? Lucas: Ceptology charges roughly $1,200 per store per month for the full suite — pressure sensors, RFID readers, smart shelf inserts, and the software platform. For Schnucks, with 14 stores, that's about $16,800 a month. They estimated the theft reduction saved them around $18,000 a month in shrink — so basically a nine-month payback period. Luna: That's a pretty clean ROI story. And it's a subscription model, so the retailer doesn't have to buy expensive hardware upfront. Lucas: Right. And the hardware itself is relatively simple — load cells, RFID antennas, some embedded microcontrollers. It's not exotic. The IP is in the software — the algorithms that can distinguish between a weight fluctuation caused by someone leaning on the scale versus someone bagging an unscanned item. Luna: Yeah, I imagine the false-positive tuning is where the real engineering lives. Lucas: Absolutely. Early pilots had issues with false positives — the system would flag someone putting their purse on the bagging area. They had to train the model to ignore objects that don't match any item in the inventory database. It's an ongoing refinement. Luna: It reminds me of how smart home sensors have to distinguish between a person walking and a pet. Same problem domain — just applied to retail. Lucas: Exactly. And this is where the IoT angle really shines. These sensor networks generate continuous data — weight trends, item velocity, dwell time at the self-checkout. Retailers can use that data to redesign store layouts, not just catch theft. Schnucks found that certain products were being stolen more frequently when placed near the exit — so they moved them to a back aisle. Luna: So the same sensors that prevent theft also inform merchandising. That's a double benefit. Lucas: Exactly. And that's the story that Ceptology sells — not just security, but operational intelligence. The CEO told me that their average client sees a 15 percent reduction in shrink and a 4 percent increase in inventory accuracy within the first quarter. Luna: Inventory accuracy is a huge deal for grocery margins. If you think your stock is 95 percent accurate but it's really 85 percent, you're making bad replenishment decisions. Lucas: Right. And the beauty of the IoT approach is that it's inherently granular. The system knows exactly how many cans of soup left the shelf via purchase versus via theft or damage. That's a level of visibility that traditional inventory systems can't touch. Luna: Are there any downsides? What happens if the network goes down — do the scales still work? Lucas: That's a good question. The system is designed to work in a degraded mode — if the cloud connection drops, the local controller can still process transactions and flag discrepancies. But the anti-theft alerts won't reach the associate's device until connectivity is restored. So there's a window of vulnerability. Luna: Interesting. So it's not a silver bullet, but it's a pretty elegant application of existing sensor tech to a real problem. And it doesn't require a privacy tradeoff. Lucas: Speaking of not requiring tradeoffs — this episode, like every episode of this show, is free and ad-free. And that's entirely because of listeners who choose to support us. A couple of dollars a month is genuinely what keeps these going — buy me a coffee dot com slash fexingo, if you've gotten something out of them. Luna: It's one of those things where a small contribution from even a fraction of our audience would cover everything. And it keeps the content independent. Lucas: Absolutely. Anyway — back to the sensors. One thing I found fascinating is that Schnucks is now experimenting with Ceptology's system in their 'scan and go' mobile app lanes. That's where customers scan items with their phone as they shop. Luna: Oh, that's a whole new can of worms. How do you prevent theft in a scan and go model? Lucas: Exactly. The sensors integrate with the app. As you place items in your physical cart, the system cross-references the phone scans with weight changes in the cart. If you scan a banana but put in two, the system can prompt you to re-scan — or automatically add the second one to your virtual cart. Luna: So it's like a frictionless checkout that still has a layer of physical verification. That's clever. Lucas: It is. And it means the store doesn't have to install expensive gantries or cameras. Just some load cells and RFID antennas. The IoT infrastructure is relatively cheap to retrofit. Luna: I wonder how long before we see this in more places. Do you think it'll become standard in grocery? Lucas: I think so. The privacy angle gives it a real advantage over camera-based systems, especially in places with stricter biometric laws — like Illinois, Texas, Washington. Those states have laws that restrict facial recognition in commercial settings. Sensor-based systems sidestep that entirely. Luna: So it's not just a technical choice — it's a compliance choice. Lucas: Exactly. And as more states consider biometric privacy legislation, I expect we'll see a lot more retailers opt for this kind of 'low-vision' approach. The sensors see your stuff, not your face. Luna: That's a nice closing thought. I think the key takeaway is that good IoT design doesn't have to be invasive — sometimes the simplest sensors can solve the hardest problems. Lucas: Totally. And if you want to see how the Schnucks pilot actually looked, we'll link to Ceptology's case study in the show notes. For now — that's it for this episode of Internet of Things with Fexingo.