Incrementality testing vs. multi-touch attribution for budget cuts: a tale of two answers

Incrementality testing vs. multi-touch attribution for budget cuts: a tale of two answers

You need to cut 20% of spend. Should you trust your MTA dashboard's lowest-credit channel, or run a lift test? They frequently disagree — and the disagreement is the lesson.

Why they conflict
Multi-touch attribution assigns fractional credit based on presence in converting paths. Incrementality measures causal lift via a holdout. A channel can score high in MTA yet near-zero in a lift test — branded search and retargeting are the classic offenders, because they sit on paths of users who would have converted anyway. MTA sees correlation; the lift test sees causation.

The asymmetry of the mistake
If you cut the channel MTA ranks lowest, you might cut a true demand-generator that simply rarely appears as the credited touch. If you cut the channel incrementality ranks lowest, you cut spend that demonstrably wasn't moving conversions. The second error is recoverable; the first quietly starves your funnel's top.

What the studies show
Replicated platform experiments (Facebook, Google) repeatedly find observational attribution overstates the lift of high-intent, bottom-funnel channels by multiples versus randomized holdouts.

Bottom line for practitioners: For budget cuts specifically, trust incrementality over MTA — you are making a causal decision, and only experiments measure causation. Use MTA to generate hypotheses about which channels to test, then let the holdout decide. Cutting on attribution credit alone is how teams accidentally defund the channels doing the real work.