We reviewed each step in the customers journey to identify a conversion point that would be more closely correlated with orders but provide a high level of data volume for machine learning to perform its best.
Once we identified that KPI (account signups), we worked with the client to push that data back into Google as an in-App event. When the new signup conversion metric reached enough volume we gradually changed the campaigns over to a tCPA objective based upon cost per signup. Each step was monitored to determine the impact on installs and orders. Then incrementally, tCPA bids were reduced 20% incrementally until the clients cost/order goal was achieved.