Viewability Work That Starts With Your Unmeasured Rate

Ask for the campaign's unmeasured rate before you ask for its viewability rate. On the display buys we inherit, between 11 and 24 percent of impressions come back with no viewability measurement at all, and when those are counted as not viewable, which is how a viewable-impression guarantee has to be counted, a campaign reported at 72 percent lands closer to 57. Most viewability programmes optimise a figure calculated on the measured subset only. That is why they improve on the dashboard and change nothing about whether anyone saw the advertisement. The three things that move the real number: Measurement coverage comes first. In-app inventory without an Open Measurement SDK integration cannot be measured at all, and a meaningful share of Spanish-language app supply still is not integrated. That is a buying decision, not a measurement problem. Second, format and placement geometry: a 300x250 in the third slot below the fold on a slow mobile connection loses on time in view to a 320x50 anchor, and both lose to a 300x600 in the article rail. Third, when the impression is counted. Counting at ad server response rather than at render inflates the denominator on connections where median load sits above four seconds, which describes most of the mobile web in this region. What we refuse: We will not buy a viewability guarantee as a product. A seventy percent guaranteed deal is a filter applied after the auction. You pay a premium for the seller to discard the impressions you did not want, and you learn nothing about which placements deserve your money. We will not set a pre-bid predicted viewability floor above sixty percent on open exchange, because past that threshold the surviving inventory tilts hard toward auto-refreshing anchor units and made-for-advertising templates that are viewable and worthless. And we will not report a viewability figure that quietly excludes unmeasured impressions, even where a holding company reporting template asks for exactly that. How the buy gets rebuilt: We pull in-view duration distributions rather than a binary viewable flag, from IAS or DoubleVerify impression-level exports joined to DSP logs, and read the shape. Auto-refresh appears as a spike at the refresh interval. Slot stacking appears as several impressions sharing an in-view start timestamp inside one page view. From there we reset the format and placement mix, move floors from campaign level to placement level, and cut what the distribution cannot defend. Changes go in one at a time with a two-week read on each, because five simultaneous changes teach you nothing about any of them. What you receive: A coverage report naming every seller and app bundle where measurement is absent and what replacing them costs in reach. A placement-level table of time in view. The revised buy, made in your seat by us or handed to your traders as a change list, your call. A monthly read for the length of the engagement. Who should not book this: Advertisers whose objective is the viewability line in an agency scorecard. We will lower that number in month one by counting honestly, and if that is a political problem this is the wrong engagement for you. Campaigns running entirely on connected TV or inside social platforms, where the constraints differ and this work does not transfer. Teams without impression-level access from their verification vendor; the aggregate export is not enough, and we would rather you spend the upgrade fee than pay us to guess at the shape of a distribution we cannot see.