For six weeks, the promo endcap in a 4,200 sq ft convenience store did nothing. The supplier funding was real. The planogram was built correctly. The price point beat the shelf next to it by thirty cents. Comp sales on that endcap sat flat, week after week, while the same promotion lifted numbers at two sister locations.
The general manager did what most floor managers do when a display underperforms: she asked to swap it. New facing, a better shelf-talker, maybe move it six feet closer to the register.
Nobody swapped anything. Instead, someone pulled up the zone dashboard the store had been running for a different reason entirely, and looked at that one camera feed hour by hour instead of as a weekly average.
The counter didn't move
The cameras covering that store weren't installed to fix a promotion. They'd been up for months, watching the register queue and the cooler wall for a shrink problem that had nothing to do with the front endcap. The zone covering that display existed in the map from day one, tagged and timestamped, logging quietly into a dashboard nobody checked more than once a week.
That's the first thing worth naming. A flat weekly average can sit directly on top of a schedule-shaped problem and never show it. Averaging by week smooths out exactly the pattern that matters most, because what happens at 7am simply disappears once it's folded into a seven-day mean.
The count was measuring the wrong people
A store this size runs 17 tracked zones in Janus. Four of them get tagged and separated out entirely as staff paths: the register apron, the walk-in cooler door, the stockroom hatch, the aisle a stocker walks to restock the front of house. That separation exists because a raw camera count can't tell the difference between a customer and an employee doing their job in the same six square feet.
A stocker facing product on an endcap for forty seconds produces the same bounding box and the same duration as a customer reading a two-for-one sign for forty seconds. Janus ties movement patterns to shift metadata, who's on the clock and which paths they repeat all day, and strips that signal out before a number ever reaches a dashboard labeled "shopper dwell."
Once that filter got run specifically against weeks of footage from the promo zone, the shopper-only signal underneath wasn't flat. It had a shape. Nobody had drawn it yet, because nobody had asked the camera to separate who was standing there.
What the hour-by-hour read showed
Two windows stood out immediately. Between 7 and 8am, the restock cart for that endcap sat parked directly in front of it, every morning, for the entire hour: the same window when the store's actual commuter rush runs heaviest. Between 3 and 4pm, an overlapping shift handoff put the closing stocker on that same aisle facing product right as the after-school and after-work traffic peaked, before the opening cashier had finished clocking out to clear the path.
Illustrative: reconstructed from the deployment's zone data to show the shape of the shift across the day. The blended +53% lift is the verified, measured number.
Both windows meant the same thing at two different times of day: right when the largest number of people were walking past that display, an employee's body and cart were standing directly in front of it. Six weeks of flat comps had actually been measuring an obstruction.
The fix was a whiteboard
Two changes went onto the schedule. Restock moved from 7:40am to 5:45am, ahead of the doors opening to the public. The shift handoff got staggered by twenty minutes, so the incoming stocker doesn't reach that aisle until the afternoon peak has already cleared. Nobody touched the shelf. Nobody touched the sign. Nobody touched a single SKU.
The tasks didn't disappear. They moved off the two hours when the promo zone actually had an audience.
For ninety minutes some afternoons, an employee's restocking cart stood exactly where a customer would have paused.
Measured over the four weeks that followed, with the identical staff-exclusion filter applied on both sides of the comparison, genuine shopper dwell in that promo zone came in 53% higher.
The number that mattered wasn't traffic
Foot traffic through the store barely moved week over week. There was no new campaign or added signage, nothing a comp-store report would ever flag as a driver. What changed was when the floor was actually clear enough for someone to stop and look, instead of walking around a cart to get past it.
That's the part most operators don't see coming. The two decisions that fixed this were decisions about people, made without anyone standing in the aisle with a stopwatch. A camera watching an empty-looking display for six weeks read like a merchandising failure. Once staff got separated out of the count, it turned out to be a scheduling gap that happened to be standing in front of a display.
The camera feed had the answer the entire time. It just needed someone to ask which pixels belonged to the customer.