Home · Fundamentals
How foot traffic is actually measured
Every foot traffic dashboard shows the same kind of confident integer: 4,218 visits last Tuesday. The number arrives clean, no smudges, no shrug. But behind every footfall figure sits one of about five measurement methods, each counting a different thing, each wrong in its own charming way. Knowing which method made your number is the difference between using data and decorating opinions with it.
Phone-based panels: the big maps with soft edges
Most third-party "visits to any store in the country" data comes from location signals collected via apps on a sample of phones, extrapolated to the population. Strengths: coverage (any location, including your competitor's), history, catchment and cross-visit analysis nothing else can do. Weaknesses, and they're structural: it's a sample with demographic skews; GPS in dense urban canyons or indoor malls drifts badly enough to confuse adjacent storefronts; visit definitions (how long, how close) vary by vendor and quietly change; and every privacy update from the phone platforms shrinks the panel further — a trendline that only points one way. Treat panel data as excellent for comparison and pattern — is this corner busier than that one, weekdays vs weekends — and as soft for absolute counts. The integer is modelled, whatever the dashboard's font implies.
Door counters: small truth, honestly told
Beam sensors and overhead 3D counters at an entrance count actual bodies crossing an actual threshold, all of them, all day. This is the gold standard for your own location — cheap, continuous, and consistent. Known sins: staff inflate counts, double doors and prams confuse cheap sensors, deliveries spike mornings. None of that matters much in practice because the bias is constant, and constant bias cancels in comparisons — week vs week, store vs store with the same hardware. If you operate a physical site and don't have a counter, that's the first hundred-odd dollars to spend in this entire field.
Wifi and bluetooth sniffing: clever, decaying
Counting phones that probe for networks once promised passive, cheap presence data. Then the platforms started randomising MAC addresses, precisely to stop this kind of counting, and accuracy has been eroding ever since. It can still indicate relative busyness and rough dwell, but as a counting method it's living on borrowed time and vendor adjustments. Useful corridor-level thermometer; weak foundation for decisions with money on them.
Cameras: rich, local and accountable
Modern vision systems count, track paths, and estimate dwell within a space — by far the richest in-venue picture, and the basis of most "conversion funnel" analytics in retail. Two costs: privacy obligations (signage, retention policies, in some jurisdictions consent regimes — do this properly or not at all) and field-of-view truth: a camera counts its cone, not your store. Good deployments map coverage explicitly; bad ones quietly count the doormat and call it the shop.
And the clipboard
Manual counts are mocked until someone needs ground truth. An hour of human counting, repeated across representative dayparts, is how every other method gets validated — and for a one-off decision like a lease, a few counted hours plus a panel comparison beats either alone. The clipboard scales terribly and lies the least.
The one-question audit
Whatever number you're handed, ask: counted, sampled, or inferred? Counted (door counter, camera, clipboard) — trust locally, mind the hardware quirks. Sampled (phone panels) — trust the patterns, not the integers. Inferred (wifi, models) — treat as weather forecast. Vendors who answer that question crisply tend to sell the better data; vendors who answer with a deck about AI tend not to.