Numbers you can defend in the debrief.
Every metric Campaign Vision reports has a fixed definition, a stated basis, and validation behind it. Here's exactly what the numbers mean — and the honesty rules they follow.
What each metric means, precisely.
Impressions
Gross exposures — every person who appeared in a camera view, counted per camera. The top of the funnel, comparable to ad impressions.
Reach
Unique people — impressions de-duplicated within and across cameras using anonymous appearance matching. Always lower than impressions; the ratio between them is your frequency.
Visitors
People who actually entered the activation footprint, as opposed to passing through the camera's view.
Walk-throughs, browsers & engagers
Visitors segmented by behavior: walk-throughs (under 3 seconds), browsers (3–30 seconds), and engagers (30+ seconds in a zone). “Engagement rate” on our dashboards means engagers as a share of visitors — a defined threshold, not a vibe.
Dwell time
How long visitors stayed — reported as a distribution (typical, 10th and 90th percentile), not just an average one outlier can distort. Visits cut short by tracking loss or event close are excluded from dwell stats rather than counted as short stays.
Zone metrics
Visits, dwell, and peaks per zone you draw on the camera view — the same definitions applied per zone, so zones are comparable to each other and across events.
The rules the numbers follow.
Measured vs. estimated, always labeled
Every number in the product is tagged as measured (computed from what the cameras saw) or estimated (derived from a model, like media value or the attention signal). The AI report is only allowed to claim what was measured.
Validated against ground truth
We routinely compare the system's output against hand-counted real footage across scene types, and that validation work continues with every release. Ask in a demo and we'll walk you through the current results — including where the system is strongest and where it's hardest.
No universal accuracy claim
Counting accuracy depends on the scene: camera angle, crowd density, lighting. A vendor quoting one accuracy number for every condition is quoting a brochure, not a measurement. We'd rather tell you a number's limits than inflate it.
Capabilities stay capabilities
Cross-camera journeys, the attention signal, repeat-visit indication, and optional demographic estimates are offered as capabilities for richer insight — directional reads, clearly framed and labeled estimated, never dressed up as guaranteed per-person facts.
What a measurement actually looks like.
Measurement questions
How accurate is it?
Honest answer: it depends on the scene, and we'll show you rather than quote a number. Detection, impressions, dwell, and zone metrics are the most robust; de-duplicating unique people gets harder as crowds get denser. In a demo we share our current real-footage validation so you can see exactly how each metric behaves in conditions like yours.
Why don't you publish a single accuracy percentage?
Because one number for all conditions would be marketing, not measurement. Accuracy varies with camera placement, density, and lighting — so we validate per scene type and label every metric's basis in the product. If a competitor's brochure says “up to 99%”, ask them: measured on what footage, and what does it drop to on a packed floor?
Can we compare numbers across events and agencies?
Yes — that's the point of fixed definitions. An engager means 30+ seconds everywhere, dwell is computed the same way at every event, so two activations measured by Campaign Vision are genuinely comparable, regardless of who produced them.
What happens with very small crowds?
Aggregates stay aggregates: heatmap cells with fewer than three people are suppressed automatically, so a quiet moment never exposes an individual. Metrics for tiny samples are shown with appropriate caution rather than dressed up as trends.
Where do the AI report's claims come from?
Only from the event's measured data. The report generator is constrained to the metrics on the dashboard — it narrates what was measured and is not permitted to invent or extrapolate numbers.
Bring your hardest measurement questions.
Book a demo and we'll walk through the definitions, the validation, and how each metric behaves on an event like yours — limits included.