Marketing attribution for CMOs who have to defend the budget

Every attribution model is wrong. The useful ones are wrong in a way you can inspect. If the CFO cannot see how a channel was credited, the credit will not survive the next budget round.

The attribution conversation usually starts in the wrong place: which model. First-touch, last-touch, linear, data-driven, whichever the ad platform prefers this quarter. The model is the last decision, not the first. Before it come three questions a board will actually ask, and most marketing reporting cannot answer any of them.

The three questions

1. How much of the spend is measured at all? Not modeled, not assumed. Measured. In our audits the answer is rarely above four-fifths, and the unmeasured fifth is where the arguments live. Start by sizing the measurement gap honestly.
2. What would have happened without the spend? This is incrementality, and it is the only attribution question a finance team truly cares about. Platform-reported conversions answer a different question: which conversions did the platform touch. Those are not the same number and the difference is usually large.
3. Can someone outside marketing reproduce the number? If the credit for a channel depends on a setting inside a vendor dashboard, the number belongs to the vendor. A board-grade attribution figure has a source, a method, and a confidence, and it can be recomputed by someone who did not build it.

Rebuilding attribution on evidence

The approach we use has four layers, and the order matters.

Layer one: the outside-in baseline

Before any internal data, score the brand's demand position from public signals: share of search against peers, AI-answer citation share, review velocity, direct-traffic proxies. These are independent of your tracking setup, which means they cannot be broken by a consent banner or a cookie change, and they move before revenue does. They are the control group your internal data lacks.

Layer two: the measured floor

Count only what is deterministically tracked, end to end, with the gaps stated. This number will be smaller than what the platforms report. That is the point. It is the floor you can defend without a caveat.

Layer three: modeled lift, labelled as modeled

Geo tests, holdouts, and marketing-mix models fill the space above the floor. Each figure carries its evidence tier: measured, modeled, or inferred. A board will accept a modeled number if it is labelled as one and the method is stated. It will not accept a modeled number presented as measured, and it will remember when it finds out.

Layer four: the re-score

Attribution is not a report, it is a loop. Every month the outside-in signals, the measured floor and the modeled lift are re-scored against the same baseline. Movement in the outside-in layer that is not explained by spend is the most interesting finding in the pack, because it points at something the plan missed.

What to stop reporting

• Platform-attributed revenue as a headline. Report it, but as the platform's claim, in its own column, next to the measured floor.
• Blended ROAS without a spend split. One good channel hides three bad ones.
• Any metric without last quarter's value beside it. A number without a baseline is decoration.

The three numbers a board will accept

Measured revenue with the gap stated. Modeled incremental lift with the method stated. Outside-in demand position against named peers, dated. Present those three, in that order, every quarter, and the attribution argument stops being about the model. It becomes a conversation about what to do next, which is the conversation a CMO actually wants to be in.