The error nobody notices and everybody pays for
A store with eight staff, each crossing the entrance a handful of times per shift for deliveries, breaks, and errands, generates dozens of entrance events per day that are not visitors. In a quiet branch that inflation can be a meaningful share of measured footfall, and because it lands in the denominator, it quietly depresses conversion rate, distorts the traffic curve that staffing decisions read, and does it consistently enough to look like a real pattern. This is the most under-discussed accuracy problem in counting, and it is not a sensor defect: the sensor counted correctly, it counted the wrong people.
How do you exclude staff from people counting data?
Four approaches, in ascending order of reliability. Zone logic: a dedicated staff entrance or a back-of-house threshold that is counted separately from the customer door. Schedule logic: excluding periods before opening and after closing, which removes the easy cases and none of the mid-shift movements. Movement-pattern logic: identifying the signatures staff produce (repeated short round trips between the same two zones, movements against the dominant flow at times of day customers do not) without identifying individuals. And explicit device or tag association, which works but introduces per-employee identification and the co-determination conversation that comes with it. Most deployments should combine the first three and avoid the fourth.
Why the fourth option is usually the wrong one
Tagging staff solves the arithmetic and creates a governance problem: the moment a system can distinguish employee movements individually, it can in principle report on them, and that possibility is what a works council reacts to, regardless of intent. The pattern-based approach keeps the system incapable of per-person reporting while still removing most of the distortion, which is why workplace deployments should reach for it first. The wider workplace measurement posture is in office occupancy analytics, and the clause that makes the commitment contractual is in contract privacy clauses.
What it costs to ignore
Three downstream failures trace back to unfiltered staff movements. Conversion rates read low across the estate, which sends teams hunting for a merchandising problem that does not exist. Small-format and low-traffic sites look worse than they are, because the fixed staff movement count is a larger share of their smaller total, which can distort a store-ranking exercise badly enough to affect decisions about the sites themselves. And the traffic curve gains phantom bumps at shift changes and delivery windows, which is exactly when scheduling models will helpfully add labor.
Verify it, do not assume it
Staff exclusion is a claim to test, not a feature to tick. The method is straightforward: during a manual ground-truth count, record staff crossings separately, then compare the system's visitor figure against the visitor-only manual figure rather than the all-crossings figure. If the vendor's number matches your all-crossings count, exclusion is not working, however the interface is labelled. Fold this into the standing procedure in accuracy test methodology, specify it explicitly per people counting specifications, and make it a line in the acceptance test. It belongs on the same list as the other quiet distortions in people counting accuracy factors and group entry counting.
---