Everything you learned in the Smart Bidding module applies here: the target is a contract, the budget is the master key, and the system needs volume to learn. What is specific to apps is the ladder: which rung to start on and when to climb.
The rungs
| Rung | What you ask for | When to use it |
|---|---|---|
| Cost per install (tCPI) | Installs at the target price | At launch, or while there are no reliable in-app events |
| Cost per in-app action (tCPA) | That the action YOU chose happens at the target price | When that action happens often and is measured well |
| Return (tROAS) | Revenue per euro invested | When there is purchase history and the amount arrives |
Public guidance on climbing is sensible and matches the logic of the Smart Bidding module: move from install to action when the action happens on the order of ten times a day in the campaign and is measured reliably; and to return after a solid month of consistent purchase data with values. Climbing too early is the classic mistake: you ask the system to optimise a signal it hardly ever sees.
Minimum budgets: the small print nobody reads
The most quoted public references: 50 × the target cost per install for install campaigns and 10 × the target CPA for action campaigns. They are not whims: they are the translation into money of the event volume the engine needs to predict. If your budget doesn't reach the rung you want, step down (back to installs) instead of holding a target the system will never be able to learn.
How to move targets
Exactly as everywhere else in Google Ads, and for the same reason:
- Small changes (the usual reference is no more than ~20% at once) and spaced out (every two or three days at least).
- One lever at a time: budget or target, not both.
- After a large change, one to two weeks of relearning with volatile results: count on it before touching anything.
- If the system can't reach the target and the campaign shrinks, don't insist by squeezing: relax the target to what the campaign is actually achieving and squeeze again slowly (the "punished campaign" manoeuvre from the Smart Bidding module).
An app-specific nuance: new versus returning
When you optimise for actions, who performs them matters: some campaigns chase the action in new users (those who just installed) and others — engagement campaigns — chase it in users you already had (lesson 8). Keep those two goals in separate campaigns or you will end up paying to reactivate people who were already yours and calling it acquisition.
💡 Ninja trick: set the target with the business calculator, not the channel's. If a purchasing user leaves €40 of margin and one in ten installers purchases, your ceiling per install is €4 — and that, not the industry CPI, is the number to defend.
⚠️ Pitfall: chasing the lowest possible CPI. Cost per install can be crushed with traffic that never opens the app again. It is the exact equivalent of the cheap lead that never buys: lesson 10 is about precisely that.
What you should remember
- Ladder: install → in-app action → return, climbing only with volume and solid measurement.
- Budget references: 50× tCPI and 10× tCPA.
- Small, spaced changes, one at a time; 1-2 weeks of learning after each big change.
- New and returning users, in separate campaigns.