Every brand has a return figure. Rather fewer can say what share of it would have arrived with no promotion at all. For a mature brand that share can be most of the volume, and it is often the largest thing standing between a reported number and a decision.
Illustrative example
Both figures are correct; they answer different questions. The first describes what converted; the second describes what depended on the spend. Budget conversations are better served by the second.
Why the channel ranking changes
Channels that look strongest under last-touch reporting tend to sit closest to a decision already made: branded search, calls to prescribers already writing, and retargeting of people who arrived deliberately. They convert efficiently, which is worth having. It is simply a different job from creating the demand in the first place.
Both jobs are worth funding. They are easier to fund well once you can see which is which.
Checks worth running on your own numbers
In roughly this order
- Hold periods back before fitting Predict six to eight recent periods cold. In-sample fit improves with every term you add; out-of-sample error does not.
- Put baseline share beside the return Show them on the same page every time. It is a one-line change to a report, and it changes the conversation that follows.
- Flag channels that move together Two channels flighting in lockstep cannot be separated by any estimator. Reporting them jointly is more useful than splitting them.
- Hold one matched region flat Read it at twelve weeks. That read becomes the opening evidence for the next decision.
One limit: this estimates average response across the period modelled. It says nothing about whether the same response holds after a competitor launch or a formulary change, which is exactly why the held-back region matters more than the model.