Rewarded and Offerwall Buying Judged on D7 Payer ROAS

We do not report cost per install. The first line of every report from this desk is cost per D7 retained payer, and on accounts we inherit the gap between that figure and the blended install cost is usually a factor of twenty or more. Rewarded and offerwall traffic installs cheaply and retains badly. Everything below exists to stop the second half of that sentence becoming your problem. Before any spend: Post-install events have to reach your attribution provider, whether that is AppsFlyer, Adjust or Singular. Installs alone are not a signal anyone can buy against. If tutorial completion, a session on day two and first purchase are not firing, the opening fortnight goes on fixing that and no media runs. This is an unpopular position to take in a first meeting. It is also the difference between buying an audience and buying a number. On iOS you receive one coarse signal per install, so the conversion value schema decides what can be learned at all. We build it around whichever early event predicts payer status in your own historical data, then check it survives crowd anonymity thresholds at your actual install volume, which is where most schemas quietly stop returning anything useful. On Android, install referrer plus server-side event postbacks. Where we buy, and the one condition: AppLovin MAX, ironSource LevelPlay, Unity, Liftoff, and the offerwall supply behind them. The condition is bundle-level reporting. A partner returning a single blended rewarded line cannot be blocklisted inside, and blocklisting is most of this job. The distance between the best and worst deciles of publishers on the same network, measured on retained payers rather than installs, is not a rounding error. We check app-ads.txt on every bundle taking real spend. Fraud, treated as arithmetic: Click-to-install time distributions catch click flooding. Install timestamp clustering catches farms. Network-claimed installs get compared against attributed installs weekly. Disputes go out with the evidence attached and a date on them, and recovery usually arrives as credit. This is routine and it is not optional. Incrementality: Rewarded audiences overlap heavily with your organic base and last-touch attribution will take the credit without being asked. Before anything scales, we run a geo holdout or a paused-cell test and size the real lift. Where the lift is not there, we say so and the budget goes back rather than into a fourth network. Advertisers outside gaming: Opt-in attention formats work for a product with a free tier and a real activation event inside the first session. They do not work for a paid trial into a subscription, where reward-motivated signups cancel at a rate that puts the cohort underwater before month two. We will not run offerwall against a trial funnel, and that refusal has cost us work. What is not included: We do not produce video creative. We write the brief, run the tests and read the results. No app store optimisation. No buying on Meta, Google or TikTok, which is a different auction run properly by different people. We specify SDK and event work; your engineers implement it. And we take no rebate, discount share or incentive from any network, so the recommendation is the one we would make spending our own money. Who this is not for: Apps with nothing worth optimising toward after the install. Teams needing install volume for a board slide in six weeks. And anyone whose media plan already assumes a rebate-funded rate, because we cannot match a price the network is paying for.