Attribution reports are models of influence, not a complete record of causality. Privacy controls, consent choices and platform-level aggregation reduce what can be observed at user level. Useful analysis acknowledges that boundary.
Separate observed from modelled
Label direct app events, attributed events and modelled estimates distinctly. Combining them without explanation creates false precision and makes changes in methodology look like changes in performance.
Use multiple decision signals
Read attributed conversions alongside blended acquisition cost, incrementality tests where feasible, regional trends and cohort quality. Agreement among signals raises confidence; disagreement is a prompt to investigate.
Match precision to the decision
Budget choices rarely require a perfect customer journey. They require evidence strong enough to prefer one action over another, with a clear plan to learn from the result.