Attribution stopped working and nobody told the budget

The reports still balance. That is the problem. A performance dashboard will reconcile perfectly against a model that stopped describing reality some time in the last two years, and nothing in it will say so.

Three things happened at once and none of them announced itself. Consent rates fell far enough that a meaningful share of sessions are now modelled rather than measured. The large platforms moved more of the journey inside their own walls, where the only account of what happened is theirs. And the click itself became a weaker signal as more of the deciding moved to places that do not produce one.

The symptom is agreement, not disagreement

When measurement breaks noisily, somebody notices. This broke quietly, and the tell is that every channel now reports doing slightly better than the business did. Add up the platform-reported conversions across a mid-sized New Zealand account and you will routinely clear the actual order count by a third.

The temptation is to pick the source you trust and carry on. That reintroduces the same error with more confidence behind it.

What survives

• Incrementality over attribution. A geo holdout is blunt, cheap and answers the only question that matters, which is what would have happened anyway.
• One reconciled number the business recognises — orders, or revenue, from the system of record — as the denominator for everything.
• Platform-reported figures kept as directional, and labelled as such in the report rather than in a footnote nobody reads.
• A modelled share disclosed on the face of the number. If 38% of a conversion count is estimated, the report should say 38%.

— Harper Digital, Auckland

New Zealand is a useful place to notice this, because the market is small enough that the actual business numbers are knowable. In a larger market the gap hides inside the scale for years.