The structural problem
Multiple channels touch the same customer, each measures independently, each claims the sale. Add the reports together and you have more revenue than your bank does.
What works in practice
- Reconcile to the store. Total ad spend against total revenue is crude and honest.
- Watch blended CAC — all marketing spend divided by new customers.
- Use holdout tests where you can: pause a channel or a campaign and watch total revenue, not campaign revenue.
What does not
Chasing a perfect attribution model. The effort is better spent making the decisions that are robust to measurement error.