The attribution you have is not the attribution you think
Most attribution models answer a question nobody asked. Here is the one they were supposed to.
By T.J. Muro
Published
Updated
Open any analytics tool and you will find an answer. The answer will be confident. The answer will be wrong in a specific, repeatable way, and the way it is wrong has very little to do with the math.
Attribution, as practiced, answers the question: given the events we managed to record, which channel touched this conversion last? The question the business asked was: which spend caused this customer to exist?
The gap
These are not the same question. The first is a database query. The second is a counterfactual. Counterfactuals are hard. Database queries are not. So we ship the database query and call it attribution.
The gap shows up in three places: missing UTMs, cross-device sessions, and the long tail of view-through. You can patch any one of these. Patching all three at once is the work.
What changes when you measure the right thing
Campaigns that looked like winners stop looking like winners. Campaigns that looked dead turn out to be carrying the brand. The CFO stops asking why the dashboard disagrees with the bank statement. That last part is the one that matters.