Most companies that watch retail cameras with AI stop at one job: catching theft. It is the obvious place to start: the loss is concrete, the return is easy to measure, and the cameras are already pointed at the problem. Argus started there too.
But once you have a real-time data layer that reads the floor, the registers, the shelves, and the cash, loss prevention turns out to be only the first thing you can do with it. The same loop that catches a concealed item can count the shelf it came from, reconcile the drawer it was paid into, answer the phone about whether you still have it in stock, and forecast how many you will sell next week. Loss prevention was the entry point. The operating system is the destination.
Argus runs as five agents on one shared loop. Each owns one part of the store. What one sees, all five see.
Operations: watching the floor
The operations agent is where Argus began. It reads every camera the store already has and catches shrink the moment it happens: bag-to-coat motions in the aisle, mis-scans and sweethearting at the register, concealment near the high-value SKUs. Every flag arrives with the SKU, the dollar value, and a short clip, not a vague “something looked off,” but a specific, reviewable event.
The shift this agent makes is from reactive to proactive. Traditional camera systems document loss after it happens. This one surfaces it while there is still a shift to protect, an associate to alert, and an item on the shelf. It is the foundation of the Revenue Integrity Score: the running answer to how much of what left your store was actually paid for.
Inventory: counting the shelves
The inventory agent counts every SKU on every shelf, continuously, and joins what the camera sees moving to what the POS says sold. That join is the quiet source of most shrink. When the camera shows four packs leaving the cooler and the register only rang two, the gap is not a mystery. It is loss absorbed into “movement,” and the agent surfaces it instead of burying it.
It also reconciles vendor deliveries tag by tag, flags planogram drift, and drafts a reorder for every short so the operator approves rather than authors. Stockouts surface the moment they happen, not when a customer points at an empty shelf.
Customer: answering the phone
The customer agent answers every inbound call and cross-checks the inventory agent and the schedule so the answers are live and accurate. A caller asking whether you have a product in stock gets a real answer, in seconds, because the agent can see the shelf. Hours, location, lottery questions, basic service: handled like a clerk would handle them, not like a phone tree.
This is what it looks like when the loop is shared rather than siloed: the agent that talks to customers is reading the same shelf data as the agent that counts inventory. The customer never hears “let me check” followed by silence, because the check already happened.
Money: reconciling the cash
The money agent watches the cash drawer in real time and reconciles it against the camera and the POS. A drawer that comes up short gets joined to the void footage and the cashier on shift before the deposit goes out, not after the monthly count finds a hole. It routes payments and flags anomalies the instant they break the pattern.
This agent feeds the Cash Anomaly Index: suspicious cash movements ranked, each one carrying the fourteen seconds of footage that explain it. Nothing slips between counted and deposited, because the gap between those two moments is exactly where the loop is watching.
Team: the workforce signal
The team signal reads the loop for what payroll structurally cannot see. Clocked-in hours are easy to measure; time actually on the floor is not, unless something is crossing the schedule against the cameras and the registers. Ghost shifts, drawer variance that follows a person rather than a register, voids that cluster on one cashier: these are patterns in the relationship between LABOR and the rest of the loop.
That is the basis of the Workforce Honesty Score: not surveillance for its own sake, but the small number of patterns that distinguish an honest, well-run shift from one quietly costing the store. The point is to give the operator the signal a single trustworthy manager would catch if they could be everywhere at once.
The forecasting layer above them
Above the four operating agents sits the forecasting layer. It reads the closed-loop data the others produce (demand by SKU, traffic by hour, cash position, labor efficiency) and projects thirty to ninety days out. That projection becomes a drafted schedule, a reorder budget, and an honest read on capital, priced against real revenue rather than a guess.
Forecasting only works because the agents beneath it produce clean, joined data. A forecast built on a single stream is a guess with a chart attached. A forecast built on the closed loop is the store telling you what it is about to do.
One score for the whole store
Five agents would be five more dashboards if they did not roll up. They do. The composite is Store Operating Health: one number per location that folds integrity, cash, workforce, and operations into a single read. An operator with two hundred stores does not open two hundred apps. They open the handful trending red and trust the rest.
That is the whole promise of an operating system: not more tools, but fewer decisions, the right ones, surfaced at the right time, on infrastructure you already own. Argus is in private beta with gas station, convenience, and grocery operators. If you want the five agents on your stores, talk to us.