App Retargeting Where the Attribution Still Works

On Android, app retargeting is a deterministic buy against GAID and works the way it always did. On iOS it largely does not, and the honest answer is that unless your ATT opt-in rate is above roughly 30 percent, or you are running AdAttributionKit re-engagement postbacks on iOS 18 and later, you are buying reach you cannot attribute. We build Android-first programmes for that reason and treat iOS as a separate question with a separate and smaller budget. The floor we will not go below: Fifty thousand matched, addressable devices in the audience. Below that the holdout group is too small for the difference between exposed and held-back users to resolve, which means you would be paying for a campaign whose effect you can guess at but not measure. If your audience is smaller, the answer is acquisition or lifecycle messaging rather than retargeting, and we will say so before an insertion order exists. Audience architecture: The most common thing we are asked to fix is one audience containing every monthly active user. That is not retargeting, it is buying your own organic returns. We rebuild the feed around intent that has expired. On a property portal: saved-search users with no enquiry in 14 days and a new matching listing to show them. On a marketplace: cart and checkout abandoners split by category value, plus sellers who listed once and never listed again. On a pet commerce app: the repeat food order window, which is a known interval per SKU and makes the timing decision for you. Each segment carries its own frequency cap and its own deep link. Deep links, where these programmes actually die: A retargeting click that lands on the app's home screen has thrown away the reason for the campaign. We implement and test Universal Links and Android App Links per segment through AppsFlyer OneLink or Branch, including the deferred case where the app has been uninstalled, and we test on real devices across every OS version you support before spend starts. It is unglamorous, and it is where most of the recoverable value in a retargeting programme sits. Incrementality rather than attributed conversions: Every programme runs a randomised holdout, normally 10 percent of the audience, inside the DSP where that is supported. Remerge and Moloco both support it. Where it is not supported, we run a matched geo holdout instead. The number we report is the difference between the two groups. Attributed conversions appear beside it and are always larger, because a retargeting platform will happily claim a purchase the user was going to make regardless. Not included: Push, email and in-app messaging. Those are lifecycle work with a different vendor stack and different economics, and we do not want to be the party judging whether the paid channel or the free one deserves the credit for the same return. Also excluded: web retargeting, creative production beyond adapting existing assets, and CRM segmentation for anything other than the ad audience feed. Who this is not for: Apps whose stated KPI is app opens. Teams who will not accept a holdout, since a holdout means part of the audience is deliberately not sold to and someone senior has to defend that internally. And iOS-only apps below the opt-in threshold above, where we would be taking a fee for something neither of us can measure. What you get monthly: Incremental conversions and incremental revenue by segment against the holdout, the attributed figures beside them so nobody is surprised by the gap, deep link health per segment, and the segments we are retiring because the intent behind them stopped predicting anything.