In-Stream Video Flights Graded on a Geo Holdout

An in-stream flight is worth the conversions it adds on top of what would have happened without it, and the only way to know that number is to withhold territory. Every engagement here splits your addressable footprint into test and control markets, DMAs in the United States, NUTS-3 regions across the EU, states or prefectures elsewhere, matched on twelve months of baseline conversion volume and seasonality, with control held dark for the whole flight. We ask for a quarter to a third of the footprint held back, for at least eight weeks. If that is not something your board will sign, we are the wrong team and will say so on the first call. How we read the skip: The five-second decision tells you more about a script than any pre-test will. We pull quartile data, played to 25, 50, 75 and 100 percent, by ad, audience and device every week, and re-cut at the second where the curve breaks rather than the second the storyboard defends. In energy retail the break is almost always the moment the ad names the supplier instead of the bill. The same asset holds far longer on a television screen, which is a property of the screen and not of your film, so we grade CTV and mobile delivery separately and never blend their view rates into one number. Two formats, two jobs: Non-skippable fifteen-second inventory buys a whole message from a smaller audience at a higher CPM. Skippable buys a self-selecting audience cheaply and tells you who selected. We use non-skippable where the message is dated and short: the closing week of an enrolment window, a tariff change with a statutory notice date, a service area going live. Skippable carries everything that has to earn its own attention. A plan that is entirely non-skippable because the completion rate reads well is a plan built for a slide. The exclusion pass before launch: Before the first impression we exclude at channel and app level, set inventory to the limited tier, remove made-for-kids, and strip the mobile game categories that manufacture accidental plays. Placement reports are read in week one and week three and the exclusion list reloaded. On connected TV we work from the app-level report and drop anything we cannot identify by name. Where you already hold an IAS or DoubleVerify contract we run their pre-bid segments against the buy. We do not resell verification and take no margin on it. Frequency: We plan two to three exposures per household per week and cap there. In the categories we work in, response past the fourth weekly exposure does not pay for the impression, and for a regulated utility the fifth arrives as complaint volume in your own contact centre, which costs more than the media did. What is not included: Film production, voice, music licensing and localisation. Channel and community management. Search, social and affiliate. We will write the second-by-second re-cut brief and sit in the edit. We do not run the shoot. What we will not report: Platform view-through conversions, as a headline or a footnote. Cost per view as a measure of success. Brand lift results from panels too small to separate the effect from the margin of error; we will run the study and tell you when the interval is too wide to act on. Who should not hire us: Advertisers whose entire case rests on last-click ROAS. Teams that cannot hold a control region. Anyone who needs a campaign live in under three weeks, since the holdout design alone takes ten working days.