Marketing measurement has degraded quietly and substantially. Browser pixels now miss 20–40% of conversions to ad blockers, cookie lifetime caps and mobile privacy controls, and platforms fill the gap with modelled estimates. Reported numbers can look stable while the underlying signal deteriorates, which is the worst possible failure mode — it produces confident wrong decisions.
Recovering the signal is engineering work rather than analytics work: server-side event transmission, correct deduplication, and importing the conversion that actually makes money rather than the one that's easiest to fire. Most of the value comes from that last change, not from recovering lost events.
But even perfect attribution answers the wrong question. Attribution is observational — it records what preceded a sale and cannot distinguish cause from coincidence. Establishing causation requires withholding advertising from someone and measuring the difference, which is why incrementality testing belongs in the measurement stack rather than as an occasional curiosity.
What decides outcomes here
Fix measurement before optimizing anything
Optimizing against untrustworthy data is worse than not optimizing, because it produces confident movement in an unknown direction.
Deduplicate before you celebrate
Running browser and server tracking together without a shared event_id double-counts. Teams see conversions jump 40%, celebrate, and are now reporting fiction in the other direction.
One revenue number, from your own systems
Blended MER — total revenue divided by total spend — cannot be double-counted or inflated. It's the only paid metric that hasn't degraded as tracking has.
Test brand search first
It's the easiest incrementality test to run, the most commonly over-credited channel, and the result most likely to change how you allocate.