Picture the job from the top of the chain instead of the bottom. A regional director with 214 stores across four states sits down on Monday morning. She has six systems that can tell her something about loss: the point of sale exception report, the video management system, the cash office reconciliation, the labor scheduler, the fuel controller, and a spreadsheet a predecessor built in 2019 that nobody fully understands anymore. Every one of them was built to answer a question about one store. She has 214 of them, and about ninety minutes before her first call.
This is the part of loss prevention that does not get written about much. Most of the advice assumes a store manager looking at their own store. The person who signs off on the budget is usually several levels above that, responsible for more locations than they can name from memory. Their problem is not detection. It is triage.
The arithmetic goes bad quickly
Say you have a genuinely good store dashboard, one that takes three honest minutes to read and understand. Three minutes across 214 stores is just under eleven hours. Run that weekly and you have created a full time job that produces no decisions, only a reading.
So nobody does it. What happens instead is that the director looks at the stores she already worries about and assumes the rest are fine. Usually they are. But a store that quietly starts slipping is, almost by definition, one nobody was watching, because a store already on the watch list is getting attention and a store that began drifting three weeks ago is not.
The number of operators in this position is larger than it looks. NACS counted 151,975 convenience stores in the United States as of December 31, 2025. Companies running ten stores or fewer own 63% of them, which is the statistic everyone quotes. The one that matters here is the other end: companies with 501 or more stores own 33,810 locations, and the cutoff to make the NACS Top 100 list in 2026 was just 66 stores, while the largest operator alone runs 6,038. Between those two poles sits a wide middle of companies with fifty to five hundred locations. That is precisely the range where you have too many stores to visit and not enough scale to fund a real analytics team.
One number per store, and the honest case against it
The obvious response is to compress everything into a single score per location and sort the list. That is what we build at ARGUS, and we call it Store Operating Health. It is worth saying out loud why an experienced operator should be suspicious of it.
A composite score hides its own inputs. A store scoring 71 could have a cash problem, a receiving problem, or one register having a bad week, and the number by itself will not say which. Scores also get gamed once people work out what moves them. And any weighting is a judgment somebody made in a meeting, wearing the costume of a measurement.
Those objections are all correct. They are also all fixable, and fixing them is most of the work. Three things have to be true before one number deserves anyone's trust.
It has to open. Every score should decompose in one step into the events that produced it, and every event into the transaction, the clip, and the person on shift. A number you cannot take apart is a rumor with a decimal point.
It has to move for a reason you can name. A score that drops because Saturday was busy is noise, and noise in a ranked list is worse than no list at all, because it spends the reader's attention on a store that was fine. Every movement should trace back to something a person could actually do something about.
It has to be comparable across stores that are not comparable. A location doing $90,000 a month inside and one doing $260,000 cannot be ranked on raw dollars of variance, or the big store sits at the bottom permanently and the small store never surfaces at all. The score has to be normalized, and the normalization has to be visible, because it is the assumption most likely to be wrong.
Most of what it surfaces will be nothing, and that is the point
There is a useful number here from outside our own work. Interface Systems published a 2026 retail loss prevention benchmark built from 1.6 million monitoring events across 18,258 US retail locations and 51 brands during 2025. Of the alarm events they examined, 270,329 in all, 95% resolved as false alarms once someone looked at the video.
Read that in both directions. It does not mean the alarms were useless. It means an alarm on its own is a question, and answering it took a second system and a human being. The value delivered was mostly negative value: 95% of the time the answer was do not send anyone, nothing is happening here.
A single number per store does the same job one level up. Most of its output is permission not to look. On a Monday with 214 stores, being told with reasonable confidence that 200 of them are behaving normally is worth more than another alert. The list should be short, and it should be short most weeks. A ranking that hands you the same fifteen stores every time has stopped being a ranking and become a habit.
Why the tolerance for this is thinner in 2026
Convenience economics moved in a direction that makes slow detection expensive. At the 2026 NACS State of the Industry Summit, NACS reported that profit on a typical in store basket, after every expense, was one cent per transaction in 2024. In 2025 it fell eight cents. Every transaction inside the store lost seven cents. Labor alone runs $1.60 per in store transaction against an average basket of $7.69.
Sit with what that does to recovery math. Under 2024 economics, a single $400 problem at one store took 40,000 baskets to earn back. Under 2025 economics it does not earn back through volume at all, because the basket no longer makes money on its own. The only lever left is not losing the $400 in the first place, and that means finding it in days rather than at the quarterly count.
That is the argument for one number, and it is a modest one. It does not catch anything a careful person with unlimited time would miss. It decides where the limited time goes. Underneath it is the same join we spend most of our days on: cash against point of sale against camera against labor, continuously, running on equipment the store already owns. The score is just what that join looks like when you have to read 214 of them before a nine o'clock call.
One last thing, because it gets lost the moment loss becomes a leaderboard. A store at the bottom of the list usually does not have a thief in it. More often it has a new manager, a broken process, or a register nobody reprogrammed after a price change. Ranking is for allocating attention, not for assigning blame, and operators who confuse the two tend to find that their scores stop being reported honestly within about a quarter.
If you oversee more stores than you can walk in a month and this is roughly your Monday, we would like to hear how you handle it today. ARGUS is in private beta. You can talk to us or write to business@useargus.co.