CPI Network Buying With Sub-Publisher Level Fraud Control

The leak in CPI network buying is one level below the network. You buy from ten networks, each of them buys from a few hundred sub-publishers, and the money that disappears, disappears there. So the first condition of this engagement is unmasked sub-publisher IDs in the postback. If a network will not pass them, we do not buy from that network, and we have ended relationships with sizeable networks over exactly this. The thresholds we cut on: Click to install time is the fastest signal. A sub-publisher whose CTIT distribution piles up under 10 seconds is showing click injection. One whose mass sits past 24 hours is showing click spam against organic installs you were going to get anyway. Either pattern earns a blocklist entry before the second day of budget rather than after a month of investigation. Alongside it we watch new device rate above roughly 90 percent, more than three installs from one IP inside an hour, and install-to-first-event conversion running under half the account average. None of these proves anything alone. Two of them on the same sub-publisher is enough to stop paying. What we negotiate before any budget moves: A 30 day rejection window rather than the seven the insertion order will arrive with. Cost per install fixed in EUR or TRY with the currency exposure named and allocated. The right to blocklist without notice. Networks concede all three when the buyer is prepared to walk, and most buyers are not. This part of the work happens before media and it determines what your recourse looks like in month three, when it matters. Attribution reconciliation: We reconcile network reported installs against AppsFlyer, Adjust or Singular weekly, by sub-publisher, and file the discrepancy while it is still inside the rejection window. The usual gap runs between 3 and 8 percent and is mostly timing. A gap above 15 percent on a single sub-publisher is a conversation with the network, and we have that conversation rather than forwarding you a spreadsheet and a recommendation. Turkey, India and the volume trap: These are among the cheapest install markets available and they are where a worldwide CPI campaign quietly ends up. If Turkey is not one of your target markets, it belongs on a geo exclusion at network level and in the postback checks, because it will otherwise arrive labelled worldwide at a CPI that makes the campaign look successful to everybody who does not open the country breakdown. If Turkey is your market, we buy it from Istanbul, in lira, from local networks that global buyers have no relationship with. Not included: Incent supply of any kind, which belongs in a different engagement with different measurement and different expectations. Programmatic bidder buying, also separate. Creative production. And any structure where we are paid per install, which would make our incentives identical to the network's. Who this is not for: Apps that need scale this quarter more than they need clean traffic, because our first month usually reduces install volume. Teams whose MMP is configured with a single postback and no sub-publisher granularity, unless the first fortnight goes to fixing that. And anyone who has already decided the fraud rate is acceptable because the blended CPI looks fine. What arrives: A blocklist that is yours and travels with you, updated weekly with the reason recorded against each entry. A network by network view of spend, cost per event and rejection rate. And a short monthly recommendation, which is normally to move budget between two existing networks rather than to add an eleventh.