You cannot track your way out of this.

Every year a marketing team somewhere buys a new tool to fix attribution. Every year the picture gets blurrier anyway. This is not a procurement failure. Privacy regulation, browser changes and platforms that keep their data to themselves have removed a large slice of what used to be trackable, and it is not coming back. Past a certain point, stitching individual journeys together stops being difficult and becomes impossible.

The instinct is to buy your way out. Better tagging, server-side tracking, another dashboard promising a single source of truth. Some of that genuinely helps at the margin. None of it restores a world where you could follow one person across the internet, and pretending otherwise produces the most dangerous artefact in marketing: a confident report that is quietly wrong.

A blurry number you understand is worth more than a precise one that is fiction.

Stop asking who to credit

The useful question was never which touchpoint deserves the credit. It is whether this spend caused anything at all. That question still has an answer, and it does not require tracking anybody. Switch the channel off in one region for a month and watch what happens. Hold out a matched audience. Compare the periods honestly. It is slower and less flattering than an attribution dashboard, and it is the closest thing to truth on offer.

The other half is modelling at the portfolio level rather than the person level: looking at spend and outcomes over time to estimate what actually moves revenue. Together, a holdout test and a sensible model tell you more than any attempt to reconstruct individual journeys from whatever survived the privacy cull. Both are honest about uncertainty, which is more than can be said for a report built to survive the Monday meeting.

The organisational part is harder than the maths

The real obstacle is political, not statistical. Incrementality testing means deliberately switching off spend that somebody is measured on, and accepting a result that may embarrass a channel owner. It means telling a board the number has error bars. Most companies would rather have a wrong number with a decimal point than a right answer with a range, which is precisely how budgets stay attached to channels that stopped working two years ago.

It is the same discipline required for demand created where you cannot watch. Accept the invisible, measure the causal, and stop paying for a precision that is not real. Serious performance work in this environment is less about the tracking stack and more about the willingness to run an experiment whose result you might not like.