Skippable in-stream to learn, non-skippable to deliver

YouTube counts a view at thirty seconds, or at completion for a shorter ad, or on an interaction, which means one word covers three behaviours and I do not report it. Skippable becomes skippable at five seconds. The gap between how many people are still watching at five seconds and how many at fifteen is the most useful measurement in video advertising, it is produced by the format itself, and it costs nothing to collect. I buy skippable in-stream to generate that curve and non-skippable to act on what it says. The first five seconds as an instrument: Three openings that differ by exactly one variable — the claim, the face, or the situation — with everything after the fifth second held identical. The retention delta between them is a read on the proposition rather than on the edit. Most video testing changes four things at once and produces a winner nobody can explain, which is why the next campaign starts from zero again. Campaign structure, deliberately unfashionable: Formats stay in separate campaigns. Video Reach and Demand Gen bundle bumper, skippable and Shorts and return one blended cost, which is fine so long as nobody intends to learn anything from the flight. Separated, you can see that the six-second unit bought reach cheaply and moved nothing, and that the twenty-second skippable carried the result. Target CPM handles delivery on non-skippable. Cost per view is never an optimisation goal here, because optimising to it buys cheaper attention rather than better attention. Television screens change the brief: YouTube extended non-skippable inventory on connected TV to thirty seconds in 2023, and a thirty-second non-skippable spot is a different creative problem than a fifteen. Meanwhile the largest waste in most in-stream plans is frequency overlap between YouTube on the television and a DSP running the same audience against the same households. The cap goes where the overlap actually happens and net reach gets reported after deduplication, not before. Brand lift, conditionally: Only above the budget where cells reach significance, and only when the question is written against a decision somebody will make. Aided awareness moves and tells you nothing you can act on. A question about the specific claim under test, asked of exposed and control groups, is worth what it costs. Not part of this: Production, organic Shorts, creator work and search. I direct the openings and review cuts against the placement; I do not shoot them. I also do not accept view rate, average watch time or completion rate as reporting currency, and I do not run bumper-only plans. Six seconds is a reminder for a proposition somebody already knows, and if nobody knows it yet the reminder has nothing to remind them of. Skip me if: The brand has no consistent claim to test. The method above answers which proposition survives contact with an audience, and it assumes there are competing propositions worth separating. It also assumes you will act on the answer, including when the answer is that the expensive spot lost to the cheap one. And if a like-for-like cost per view against a previous agency's number is how this will be judged, we are not a fit — that number is a function of format mix, market and bid strategy, and it can be improved by buying attention worth less.