Walk into almost any retail org and ask who owns shrink. The answer is loss prevention. Which is exactly why shrink never gets solved.
Loss prevention is a security function. It is staffed with investigators, built around cases, and measured on incidents. So when you hand that function the shrink number, it does what its DNA tells it to do. It hunts theft. Cameras go up. Cases get locked. Guards get hired. The whole apparatus points at the perimeter.
Then the data comes back and tells a completely different story.
The variance is not where the budget is
The 2026 Appriss Retail benchmark is the cleanest read I have seen on this. $90 billion in losses due to shrink, 73% of which was preventable due to employee theft ($26 billion), inventory errors ($19 billion), operational errors ($12 billion), and organized retail crime ($9 billion).
Read that order one more time. Organized retail crime, the thing driving your security spend and every headline, is the smallest of the four buckets. The three larger buckets are not crime problems. They are data and process problems wearing a crime costume.
Employee theft sounds like a security issue until you look at the mechanism. Most employee theft occurs through POS manipulation rather than direct merchandise removal, making transaction-level analytics essential for detection. That is not a guy stuffing product in a jacket. That is a void pattern, a refund anomaly, a discount override signature sitting in your transaction log. You do not catch it with a camera. You catch it with exception analytics on POS data.
Inventory errors are pure data. Operational errors are spoilage, damage, and markdown timing, which is a forecasting and pricing discipline, not a theft event. Add those two and you are at $31 billion, more than three times the ORC number.
The industry keeps optimizing the smallest lever because that is the lever loss prevention knows how to pull.
The wrong function owns the biggest number
Here is the structural problem. Shrink is a single reported number that rolls up multiple root causes owned by different functions. Inventory accuracy belongs to supply chain and store ops. Markdown timing belongs to merchandising and pricing. POS exception handling belongs to finance and systems. But the aggregate lands on loss prevention, a function with neither the mandate nor the toolset to touch any of them.
So the org that owns the number cannot move the variance, and the orgs that can move the variance do not own the number. That gap is where the money leaks.
And the base rate is brutal. The average U.S. retailer operates at just 65% inventory accuracy. When your perpetual inventory is wrong a third of the time, everything downstream inherits the error. That level of inaccuracy cascades into stockouts, excess safety stock carrying costs averaging 20 to 30% annually, demand forecasting errors, and customer loyalty erosion.
That is the tell. Shrink is not just a line item that dents gross margin. It is a corrupting input that poisons forecasting, replenishment, and pricing across the entire chain. You cannot camera your way out of that.
Put shrink where the variance lives
The forward-deployed answer is not more hardware. It is reassigning the problem to the function that owns the data and giving that function production AI that operates on transactions, not just video.
Start with POS exception detection running continuously against every void, refund, override, and no-sale. Layer inventory reconciliation that reads receiving, cycle counts, and sales into a single accuracy signal per SKU per store. Add markdown timing models that fire before product spoils, not after it hits the clearance rack. None of that lives in a camera. All of it lives in data you already generate and mostly ignore.
The discipline sequence matters too. Effective shrink reduction starts with foundational process controls such as cycle counting, separation of duties at POS, and receiving dock discipline, before layering technology on top. Process first, then AI on top of clean process. Not AI as a substitute for it.
Now the Cost of Doing Nothing. Every quarter you keep shrink filed under security, the variance in inventory accuracy compounds silently. It shows up as phantom stockouts, inflated safety stock, and forecast error you attribute to demand volatility. You think you are holding the line with cameras. You are actually financing the gap and calling it theft.
Stop counting cameras. Start counting reconciliation cycles, exception hit rates, and inventory accuracy by SKU. Move the shrink number out of the security org and into the data and process orgs that can actually bend it.
The retailers who win the next cycle will stop treating shrink as a crime to investigate and start treating it as a variance to engineer. The costume is coming off either way.