Unity, ironSource and AppLovin buying with supply hygiene

Buying on Unity Ads, ironSource and AppLovin is not campaign management in the sense the other channels use the word. You are given a bid or a return goal, a geography, an operating system, and creative; the model does the rest. Which is exactly why the work concentrates in the two places nobody sells: supply hygiene at bundle level and the fraud read. On Android accounts I take over, the publisher bundles I end up blocking usually account for between five and fifteen per cent of spend and almost none of the retained users, and they are identifiable inside an hour if the raw install export is available. The fraud read: Click-to-install-time distributions get pulled per publisher bundle from your MMP's raw data, never from a dashboard. A healthy curve puts most installs between roughly twenty seconds and two hours. A spike inside ten seconds is click injection. A long flat tail sitting behind an implausible click volume is click flooding, which does not add installs, it takes attribution from your organic and from every other channel you are paying for. Bundles get blocked, the network receives an evidence file, and I ask for the credit. Networks settle these claims more often than buyers expect, and never for a buyer who cannot show the distribution. The return goal: Set at D7 against your own cohort figures, revised monthly, moved in single steps because each change restarts the model's exploration. One campaign per network per operating system per geography tier, never per country, so spend concentrates rather than spreading into a hundred learning phases. Raising the goal to fix a bad week is the reflex I spend most of my time arguing against. The usual cause is a creative that stopped working or a bundle mix that shifted, and both are visible before the goal needs touching at all. Creative: Playables carry this inventory. Under five megabytes, first interaction inside three seconds, and an end card showing the thing the playable just let someone do. I brief and direct playable and interactive end card production with your studio. Video-only accounts almost always leave the cheapest installs on the table, because nobody ever built the format the supply was designed around. What counts as the number: Your MMP cohort. Network-reported return on ad spend is treated as a claim and reconciled weekly, and the gap between the two is itself a diagnostic with a short list of causes, a widening gap most often meaning a supply change rather than a market one. On iOS the work runs from SKAN postbacks with the value schema fixed before spend starts. iOS and Android are reported separately rather than blended into an average that describes neither. I refuse: Percentage-of-spend fees on channels where the model spends the budget for you. Accounts where the raw MMP export is unavailable, since the fraud work without it is theatre. Network account manager recommendations passed off as strategy. And the mediation side of these same platforms. I buy here, I do not run your waterfall, and one person holding both jobs is a conflict I would not accept from somebody else. Who should not book this: Any app whose conversion is a lead form, a demo request or a professional consultation. This is games and casual entertainment inventory, the user is mid-session in something else, and considered-purchase intent is not there at any bid. Apps without in-app purchases or another measurable revenue event, since a return goal needs something to aim at. And operators wanting these networks used to reach markets where their licence does not cover them.