In one line
Serving ads to people who already know or previously installed your app, so they come back, is retargeting (also called re-engagement). Both the audience and the goal differ from new-user acquisition.
Why it always looks efficient
Open the report and retargeting CPA beats acquisition comfortably. "Put more budget here" follows naturally.
The catch is that the audience already knows you: people with the app installed but unopened, people with items in a cart. Many of them would have come back within days regardless.
When that return lands in the ad report as a conversion, performance the ad did not create gets recorded as the ad's. It is the same structure as cannibalization.
So judge it on incrementality
CPA is weak evidence in retargeting. What you need is how many would not have returned without the ad — incrementality.
Checking it is straightforward: randomly exclude part of the audience and compare return rates between the groups. If the excluded group came back at a similar rate, the campaign created less than it claims.
Do not mix it with acquisition
The two address different user states and different goals. Combined in one campaign, overall CPA looks flattering, and scaling on that number does not raise actual new users.
Split the campaigns and manage exclusion rules explicitly.
Go deeper
Designing exclusion rules and managing incremental effect is covered in the retargeting and re-engagement guide; measuring true lift with a holdout is covered in incrementality measurement.