CPI Network Buying for Regulated-Category Apps

The first decision is whether to buy at all: A paid install pays back only if the retention curve underneath it holds. For subscription apps in the categories I work in that threshold sits near 24 percent D7 and 11 percent D30; below it, network volume mostly buys a faster leak. Every engagement opens with a read of your existing cohorts against that line. If the curve says the app is not ready, I say so in week one and you pay for the read rather than for media. Which inventory I will take money to buy: Networks that return unmasked sub-publisher identifiers in the postback, with the account configured for source-level reporting before the first euro moves. Inventory brokered through a chain that resolves to a hidden bundle does not get bought here at any price, because you cannot exclude what you cannot name. That single rule removes most of the cheap end of the market. It is also why the CPIs I report tend to look worse than the ones you have been quoted, and why the D30 ROAS tends to look better. Measurement, on both platforms: On iOS the work is conversion value design before it is bidding. Sixty-four fine values and a coarse tier per postback window is a small budget of information, and most accounts spend it encoding five questions at once. I encode one, usually the activation event that predicts D30 in your own data, and design the three postback windows around it, including what happens when crowd anonymity drops you to coarse. On Android, Play Install Referrer plus deterministic attribution, with the referrer decrypt verified rather than assumed. Fraud, treated as an accounting function: Click to install time distributions per sub-publisher, read weekly. A spike under ten seconds is click injection. A flat tail past twenty-four hours is click flooding. New device ratios above the category norm, or installs arriving on a mechanical hourly rhythm, get the publisher blocked the same day with raw logs attached to the chargeback so the network cannot argue it. Recovered spend is credited to your media line, not to my fee. Regulated categories: For an app under MDR the ad creative sits inside the same claim boundary as the listing. Copy a growth team would write without a second thought, outcome language or a comparative benefit or a testimonial, is the copy that gets a Class IIa app pulled. Ad reviews here route through the same approval path as the store page, which slows the creative cycle to roughly one refresh a fortnight. Budget for that rather than be surprised by it. Polish energy and utility apps carry a comparable constraint on switching and tariff language. Not included: Creative production, SDK or MMP implementation, store listing work, and Apple Search Ads or Google App Campaigns. Those are self-attributing networks with different economics and I keep them out of this scope deliberately. Who should not buy this: Anyone without a live MMP and at least eight weeks of organic cohort data. Anyone pre-launch. Anyone who wants a guaranteed CPI, a number only ever guaranteed by somebody controlling the supply side of the trade. And anyone who wants incentivised, rewarded or offerwall installs. I do not buy them and I will not sit beside them in the same account, because they poison the retention signal every other decision depends on.