A great deal of energy goes into perfecting attribution models, on the assumption that certainty is available. It is not, and waiting for it is more expensive than deciding without it.
Contribution thinking, holdout tests and a tolerance for approximate answers will outperform a precise model built on the wrong assumptions.
The goal is not the correct number. It is a decision you would make again.