When a conversion drop is attribution, not performance

It arrives as a chart with a cliff in it. One campaign, one channel or one conversion action falls sharply while everything else in the account looks ordinary. The reflex is immediate and always the same: check the ads, check the landing page, check whether a competitor moved, and if none of that explains it, cut the budget.

The tell that it is attribution rather than performance is that the fall does not show up anywhere outside the ad platform. Revenue is flat. The booking system is flat. The phone rings as often. Only the credit moved.

Conversions do not add up across platforms, because each one is claiming credit for the same customers under its own rules. When one channel scales, it starts intercepting more of the same journeys: it sees the customer earlier, or more often, or in a place its attribution model weights heavily. The other channel is still doing the same work with the same people, and is now losing the argument about who caused it.

The same thing happens without any spend change at all. A shift in a platform's own attribution model, a change to how much modelled data it includes, or a consent change that removes some deterministic signal will all move credit between channels while the underlying business does nothing.

None of this shows up as an error. Both platforms report confidently, and their reports are internally consistent. They simply disagree with each other, and their sum disagrees with reality.

The cost is the intervention. A channel that is quietly still working gets cut because its scoreboard fell, and the real outcomes follow a month later, at which point the cut looks justified by the very decline it caused.

It also costs the reverse mistake. The channel that absorbed the credit looks like it is scaling beautifully, so it gets more budget on the strength of results it did not produce, and the account concentrates spend into the channel with the most generous attribution model rather than the most productive one.

The rule is short enough to apply every time: before treating a conversion drop as a performance problem, check the sibling channels' claims against a source-of-truth series that does not move.

Three readings settle it. If total real outcomes are flat and one channel fell, credit moved. If total real outcomes fell in step, something real happened. If total real outcomes fell more than the channel did, the problem is bigger than the campaign you were looking at.

This costs about ten minutes and it is the difference between a correct diagnosis and a confident wrong one. It is also the only version of the check that survives an account with more than one channel, which is most accounts.

It is not an argument that campaigns never underperform. They do, often. This is a rule about the order of the questions, not about the answer.

It is not incrementality testing. Comparing claims against a source of truth tells you where credit went. It does not tell you what would have happened with no spend at all, which needs a holdout or a geographic test.

It is not solved by picking one attribution model everywhere. A single model applied across platforms still allocates credit by rule rather than by cause, and the sum still will not match the business.