CTV Measurement: Attribution to Contribution

Summary

CTV underperformance in last-touch dashboards is usually a measurement artifact: the channel rarely produces a click, and the conversion often occurs later on a different device. Last-touch, multi-touch, time-decay, and position-based models depend on cross-device identity, which breaks precisely at the CTV-to-mobile boundary. Attribution assigns credit; contribution proves causality. Only incrementality testing, via geo-holdout, ghost bidding, or PSA control, shows whether conversions would have happened anyway.

With IAB forecasting $80B in US digital video spend by 2026 and linear TV below 40%, measurement readiness matters. Measure CTV in four layers: delivery (impressions, unique households, frequency, viewability), engagement (VCR, quartiles, interactive engagement), outcome (attributed installs, site visits, CPA, ROAS), and contribution (incremental lift). For IP-based view-through attribution, fire an impression URL with the CTV device IP, match same-household mobile installs within a configurable window with a 24-hour minimum, and keep deep linking disabled for CTV. Because 66% of CTV installs land within six hours, overly short lookback windows understate performance.

Before launch, define KPI hierarchy and design the incrementality test with holdout regions. In-flight, monitor delivery and completion as leading indicators, not last-touch attribution. Post-campaign, report delivery, attribution, and incrementality separately; if last-touch and geo-holdout diverge, trust the holdout. Choose tools by data sources, identity resolution, cross-device reach, and native incrementality support. An MMP fits cross-channel CTV plus mobile/web; a DSP is limited to its buying platform; pure-play measurement vendors are stronger for audience reach than install-level attribution. AppsFlyer reports CTV Attribution customers see 4x ROAS and 20% LTV lift versus last-touch alone. The budget question is contribution: did CTV drive growth that would not have occurred otherwise?

Analyst Note

This piece lands at a moment when CTV spend is outrunning the measurement frameworks meant to justify it. IAB's $80B US digital video forecast, with linear falling below 40% for the first time, is the headline number, but the more telling detail is how much of that money still gets evaluated with click-based logic on a channel that produces no clicks. What's notable here is the framing of attribution and contribution as outputs of separate mechanisms rather than competing models on one spectrum — a distinction long standard in marketing science but slow to reach operational workflows, where last-touch dashboards remain the default.

The key implication for UA teams is structural: identity at the CTV-to-mobile boundary is probabilistic by nature, via IP-based household matching, and any vendor describing deterministic resolution there is describing something else. Worth watching is whether incrementality testing, currently positioned as a premium or managed offering, becomes a baseline expectation across MMPs and DSPs alike — geo-holdout design being the one method that holds up as shared identifiers erode everywhere else.