Honest accuracy, published
What we measure, how we measure it, and exactly where the numbers stop being trustworthy.
Why this page exists
In January 2026 Oxfordshire County Council publicly withdrew its November 2025 Oxford congestion-charge footfall figures — its supplier had reweighted its panel mid-year, the figures couldn't be defended, and the headlines followed. The lesson: any organisation that buys footfall data without understanding the methodology is one supplier-side change away from the same press cycle. We publish ours.
What Crowd-Sense measures
Density and trend — how busy a place is, by zone, time, and distance band. Reliable for relative comparisons (busier than yesterday? busier than last hour? busier than last Saturday?) and operational decisions.
What it doesn't measure: absolute unique-visitor counts at long windows, return-visit behaviour, individual movement paths, or demographic attributes. Architecturally impossible.
The MAC randomisation overcount
Modern phones rotate the Wi-Fi address they broadcast every few minutes — a privacy feature, deliberately designed by Apple and Google. A Collector that respects that rotation (rather than trying to defeat it) systematically overcounts unique devices at longer time windows.
| Time window | Typical overcount | Reliable for |
|---|---|---|
| 1 minute | 1.2×–2× | Live event-time decisions, queue management |
| 5 minutes | 1.5×–3× | Short-term trend, staff allocation |
| 15 minutes | 2×–5× | Operational reporting, cleaning, HVAC |
| 30 minutes | 3×–7× | Hourly trend, capacity management |
| 60 minutes | 4×–10× | Long-window density and volume reporting |
For density and trend (the use case Crowd-Sense is designed for), the overcount factor is approximately constant for a given venue — so relative comparisons are reliable even though absolute device counts inflate.
If a vendor claims 95%+ accuracy at long windows
They're doing one of three things:
- Re-identifying randomised MACs through fingerprinting techniques that are legally contestable under UK GDPR.
- Mixing in non-Wi-Fi signals like cameras or app SDKs, and conflating them with the Wi-Fi number.
- Overclaiming.
We invite customers to ask which. The answer is informative.
Cite us properly
For cabinet papers, BID annual reviews, or anywhere our numbers will be published, here's the form of words we ask you to use:
Footfall data sourced from Crowd-Sense passive Wi-Fi sensors. Device identifiers are discarded on the sensor; only aggregate counts are transmitted. Mobile devices randomise their broadcast addresses, which causes a known overcount of unique humans at longer time windows (1.2–2× at 1-min, 2–5× at 15-min, 4–10× at 60-min). These figures should be read as device-density and trend indicators, not absolute unique-visitor counts.
The full Methodology & Accuracy PDF is available for download.