Why your camera count data is wrong (and how to fix it)

If you have ever sat in an operations review and watched two teams argue about which footfall number to trust, you already know the problem. The cameras say one thing. The gates say another. The Wi-Fi presence platform tells a third story. POS suggests something different again. The dashboard is technically working — but no one trusts it enough to act on it.

Why count data drifts across sources

Each source measures a slightly different thing, with a slightly different bias:

• Cameras count people crossing a virtual line; bias depends on optics, ceiling height and how the model handles groups, prams and tailgating.
• Gates and turnstiles count entries authorised — they will undercount visitors and overcount staff using the same lane twice.
• Wi-Fi presence samples devices; results swing with opt-in rate, MAC randomisation and how dwell is defined.
• POS counts transactions, not people — useful for conversion only when paired with a clean footfall denominator.

Every source is partially right. None of them is the truth. Without reconciliation, the dashboard inherits the disagreement.

What a single source of truth actually requires

A workable single source of truth is not a single sensor — it is a single, reconciled dataset that all sources feed and that all teams agree to act on. Building it requires:

1. A canonical model of the site (zones, lines, entry/exit definitions) that every source maps onto consistently.
2. Reconciliation rules — explicit logic for how disagreements between sources are resolved, and which source wins under which conditions.
3. A measured ground truth at every site — a periodic, audited count against which all sensors are graded.
4. A published accuracy figure on every chart, so users can see how confident the data is for the question they are asking.
5. Versioned definitions — when the model changes, the chart changes and the change is visible.

Common anti-patterns

• Picking the highest number — “the one that makes the report look best” is the fastest way to lose internal credibility.
• Averaging sources — hides bias instead of resolving it.
• Only using one source — looks clean but ignores the structural blind spots that source has at this site.
• Reporting the data without the accuracy — invites every conversation to start with “how do you know that number is right?”.

Where to start

Pick one zone, ground-truth it for two weeks, and document the reconciliation rules explicitly. Once reconciliation is working at one zone, generalising to the rest of the site is mostly mechanical — and the day teams stop arguing about which number is right is the day the dashboard starts being treated as authoritative.