Overview
Cross-screen video reach is not the sum of television, OTT, and digital platform totals. A defensible unduplicated audience requires comparable exposure rules, a shared universe, disclosed coverage, supported identity evidence, and explicit overlap treatment. The series moves from the common ruler and reporting unit to feasible Venn structures, bounds for unobserved reach, and conditional-independence models as assumption-bearing refinements. It also shows how privacy, invalid-traffic controls, quality, and auditability limit what a number can support: exposure, not attention or effectiveness. Synthetic validation results do not establish a real-campaign error rate.
Learning outcomes
- Adding television, OTT, and digital platform totals does not produce unduplicated campaign reach without compatible definitions and overlap evidence.
- A credible total requires a shared exposure basis, intended universe, coverage disclosure, reporting unit, and supported identity links.
- Model-free bounds show the reach values allowed by observed subset data before assumptions narrow them to a point estimate.
- Conditional-independence estimates can aid planning, but fit is not truth and synthetic error results do not establish real-campaign accuracy.
Keywords
Preview every chapter · Unlock full text, audio, and video from chapter 4
PlansFAQ
Can television, OTT, and digital reach numbers simply be added?
No; adding them can count the same people more than once unless the reports share definitions and have been deduplicated.
What makes a cross-screen exposure comparable?
A comparable exposure uses a common counting event, viewability basis, invalid-traffic treatment, ad or content granularity, and time rules.
How can reach be estimated when platform overlap is missing?
Use non-negative Venn-region constraints to calculate a feasible lower and upper bound for the specific unobserved subset before adding model assumptions.
When is a point estimate suitable for planning?
It is suitable as a planning scenario when its assumptions and relevant held-out validation are visible, and it remains within the model-free bounds.