Attribution vs incrementality: what to trust, when, and how to test

A practical guide to attribution vs incrementality: common attribution models, window pitfalls, how MER/marginal ROAS fit in, and how to run holdout/geo tests.

Updated 2026-01-28

Try it in a calculator

The core problem

Attribution assigns credit. Incrementality estimates causal lift (what ads caused). These are not the same, and confusing them is one of the fastest ways to scale spend into diminishing returns.

What each approach is good at

ApproachGood forCommon failure modes
Platform attributionFast feedback for optimizationOver-credits retargeting; windows differ; model changes
Analytics attributionCross-channel visibilityLast-touch bias; missing view-through; tracking gaps
MER (top-down)Alignment and sanity checksHides which channel is working; needs clean spend definition
Incrementality testsCausal truth for scaleContamination, low power, short tests vs purchase cycle

Attribution windows (why comparisons break)

  • Keep windows stable when comparing periods; changing windows changes the metric, not performance.
  • Longer windows can inflate credit (especially for retargeting); shorter windows can under-credit longer-cycle products.
  • If you must change windows, annotate and avoid comparing across the change.

How to build a measurement stack that scales

  • Start with funnel math (CPM/CTR/CVR) plus profit guardrails (break-even targets).
  • Add MER to align marketing and finance on top-down efficiency.
  • Use marginal ROAS (or incremental profit) as you scale to detect saturation.
  • Run incrementality tests when spend is meaningful or when channels fight for credit.

How to run incrementality tests (pragmatic checklist)

  • Pick a clean holdout: ensure the control truly has no exposure (avoid contamination).
  • Run long enough to cover the purchase cycle and reduce noise.
  • Measure both conversions and profit proxies (margin-aware outcomes).
  • Repeat when creatives/audiences change; incrementality is not permanent.

FAQ

Is incrementality always necessary-
Not always. Early on, attribution can be good enough for learning. Incrementality becomes more valuable as spend grows, audiences overlap, and attribution bias increases.
Can MER replace attribution-
No. MER is top-down truth and alignment, but it won't tell you what to change. Use attribution for action and MER to sanity-check.

More in paid ads

ARR waterfall: reconcile starting ARR to ending ARR (net new ARR)
Blended CAC vs paid CAC: when each is the right metric