An impression that isn't viewable, isn't from a human, or can't be attributed isn't worth paying for. This is how the industry tries to make sure.
MRC defines a viewable video impression as 50% of pixels in view for at least 2 continuous seconds. OMID standardizes how vendors measure it.
On CTV, viewability is mostly assumed (full-screen by default) but ‘device on’ is a separate question — OM SDK for CTV adds attention signals.
Pre-bid filters (TAG, GARM categories) reject the bid request before a creative ever loads. Post-bid scans flag URLs where the ad ended up.
Brand safety = avoid harmful contexts. Brand suitability = match contexts to the brand's appetite (a beer brand has different suitability than a kids' brand).
GIVT is obvious bots and crawlers — easy to filter. SIVT is sophisticated: residential proxies, hijacked devices, fake CTV apps. Detected with behavior signals.
Every major DSP runs SIVT scoring at bid time and refuses to pay on impressions tagged after-the-fact.
Linking a viewed ad to a downstream action: web visit, app install, purchase. Increasingly probabilistic post-cookie; clean rooms and MMM are filling the gap.
View-through attribution credits an impression that didn't get clicked but happened before a conversion. Sensitive to lookback window length.
CPM (price per 1,000 impressions), fill rate, VTR (view-through rate / completions), VCR (video completion rate), eCPM (normalized across deal types).
Discrepancies of 5–15% between any two systems are normal. Anything bigger and someone is reporting differently — usually the player vs the ad server.