Measurement & Attribution

Multi-Touch Attribution · The Promise and the Problem

MTA models distribute credit across the touchpoints leading up to a conversion. The math, the modeling choices (linear, time decay, U-shaped, data-driven), and why most teams are quietly retiring MTA in favor of incrementality and MMM.

Attribution. Multi-touch attribution has roots in econometric and marketing science work going back decades. Google's data-driven attribution (DDA, originally Shapley-value-based) and Facebook's attribution products popularized MTA in the digital era. This article reviews the techniques and the modern critique.

What MTA tries to do

Multi-touch attribution (MTA) distributes credit for a conversion across the touchpoints the customer encountered before converting. Instead of giving all the credit to the last click (or the first), MTA splits credit based on a model — linear, time-decay, U-shaped (W-shaped), or data-driven.

The promise: a fair view of which marketing investments deserve credit. The reality has been more complicated.

The common MTA models

Why MTA is quietly being retired

Three structural problems have eroded MTA's credibility:

  1. Privacy changes broke the data. iOS ATT, ITP, ETP, and Privacy Sandbox all reduce the user-level tracking MTA depends on. Most MTA today is operating on degraded data and producing increasingly fictional models.
  2. Dark social and offline are invisible. MTA only credits trackable touchpoints. A buyer who heard about you on a podcast, saw a billboard, and got a recommendation in a Slack DM appears to have arrived "direct" — and MTA invisibly gives all credit to whatever was clickable.
  3. Correlation isn't causation. MTA assumes the touchpoints in a converter's path caused the conversion. They may have just been present. Incrementality testing reveals the difference.
The 2026 measurement triad is replacing MTA. Media Mix Modeling for cross-channel allocation. Incrementality tests for causal validation of specific bets. Brand lift studies for upper-funnel impact. MTA remains useful for short cycles in highly trackable environments — but it's no longer the default attribution methodology for sophisticated teams.

When MTA still works

For short conversion cycles (under 7 days) in heavily-tracked digital-only journeys (e.g., a DTC purchase after seeing 3 paid social ads and an email), MTA still produces reasonable, useful credit allocation. The same model fails for B2B with 90-day cycles or DTC with significant dark-social influence.

Related on RGM

Sources & further reading
  1. Shapley, L. S. (1953). "A Value for n-Person Games." (Mathematical basis for data-driven attribution.)
  2. Google Ads Help — Data-driven attribution documentation.
  3. Group M, IAB, MMA — published critiques of MTA limitations in the privacy era.
  4. Common Thread Collective and Tinuiti — agency-side MTA retirement essays.
  5. RGM operator notes — attribution model selection 2022–2026.