Google App Campaigns rebuilt around one conversion event

Google publishes the number most App campaign accounts quietly fail: a campaign optimising for an in-app action wants roughly ten conversions a day before its bidding model is doing anything you could honestly call optimisation, and the recommended daily budget is at least ten times the target cost per action. An operator producing forty first deposits a week, or a legal-tech app producing six qualified intakes a day, does not clear that bar on the event it actually bills against. So the work here is not bid management. It is deciding which event Google is allowed to see, proving the ratio between that event and revenue, and consolidating the account until budget lands in few enough campaigns for the model to learn anything. Choosing the optimisation event: I take the earliest event in your funnel that fires often enough to clear the threshold and still moves with revenue: registration, a completed profile, a first search, a document upload. Then the ratio between that event and the paying one gets measured monthly from MMP cohorts, and the target cost per action is reset against it. When the ratio drifts, the target moves. That is the entire control loop, and it is why I will not sit inside the interface nudging numbers daily. Target return on ad spend goes on only once purchase values have been passed long enough to show a distribution rather than an average. Geography and budget: Accounts arrive split one campaign per country, each on thirty euros a day, each permanently in learning. I pool markets into two or three payback tiers, cut the campaign count hard, and let smaller markets ride inside a tier rather than starve alone. Tiers come from D30 cohort payback in your own data, never from Google's regional groupings. Country performance stays visible in reporting; it simply stops being a budget container. Assets: Five headlines at thirty characters, five descriptions at ninety, and video Google will cut into placements you cannot inspect. I brief three to five video concepts a quarter, keep what survives on installs-to-event rather than click-through, and stop there. Asset volume is not a strategy on this channel; it is a way of making the reporting unreadable. Tracking: Firebase and your MMP will both claim the same conversion when both are wired to bidding, and I have inherited accounts that spent months paying for double-counted events. One source becomes the source of truth and the other drops to reporting only. The Play install referrer has to be reaching it. On iOS, modelled conversions get reported as modelled, on their own line, not blended into a single number that lets everybody believe what suits them. What I do not do: Open a campaign per country. Hand Google the deposit event when it fires eight times a day. Edit targets more than once a week, which resets the learning period you have just paid for. Report installs. Run buying and creative as one undifferentiated line item. Or accept a mandate where someone else owns the conversion definition while I own the result. Wrong engagement if: You are Android-only in one small market, where the volume will never support this machinery and a smaller channel will serve you better. You want spend live inside a week; the tracking review takes two. Or your success metric is cost per install, in which case any buyer alive will beat my number and none of them will beat the payback. Deliverables are written. Every consolidation decision arrives with the counterfactual attached, so your analyst can disagree with me using the same data I used.