AppLovin, Unity and ironSource buying against D7 ROAS

On AppLovin's AXON and Unity LevelPlay you are not buying placements, you are training a bidder. The lever is the value you send back: a server-to-server purchase or subscription event inside twenty-four hours of install, plus a modelled value for revenue that arrives later. An account that returns only installs gets priced as though every user is worth the median, and it pays the median for as long as it runs. How the account is built: One campaign per bid goal per geo tier, with day-seven ROAS targets read off your own payback curve rather than a number the network suggests. New campaigns launch on a single creative concept so the first week of data means something, then rotate at ad level. Android and iOS are separated completely, since the signal available on each is different and mixing them produces a bid the model cannot defend. Budgets move on a weekly cadence, not daily; these bidders punish the operator who keeps flinching. Source-app work and its limit: We pull source-app-level reports, block the obvious offenders, and stop at roughly thirty exclusions per campaign. Past that point the problem is not the publisher list, it is your bid or your creative, and continuing to exclude is theatre that makes a chart look tidy while the account stays broken. We say this at the start because most incoming accounts arrive with several hundred exclusions and a belief that they are close. Three sets of numbers: The network's self-attributed figure, your MMP's figure and SKAN's postbacks will not agree, and no amount of configuration makes them. We reconcile to the MMP, report the delta every month with its direction and size, and set spend decisions against the MMP number alone. A partner whose gap widens without explanation gets its budget cut before we ask them about it. If you also sell inventory: Buying on AppLovin while monetising through MAX, or buying Unity Ads while selling through LevelPlay, is workable but it needs a wall. Your monetisation floors are commercial information about your own demand, and a buying team that reads them will bid worse, not better. We do not take both sides of that account, and if your mediation is managed in-house we will ask which of your people can see what. Not included: Mediation setup, floor price optimisation, and anything on the publishing side of your business. SDK integration and MMP implementation; we will specify what we need and review it, but your engineers write it. Creative production, which is priced separately because playables and rewarded video are their own discipline. Google Ads UAC and Meta, which are not this engagement. Who this is not for: Apps whose only reliable postback is the install. Apps in a category with no rewarded or full-screen inventory to speak of. Teams who want the account handed over with an SLA on response time rather than on decisions, and teams who intend to run the same geos through a second agency at the same time so the two can be compared; you will simply be bidding against yourself and paying for the privilege. What you are paying for: Knowing when a campaign that looks like it is recovering is actually dead, when a ROAS target should be loosened rather than tightened, and when the network's account manager is describing a product roadmap rather than something that will help you this quarter. That last one is worth more than it sounds.