Multi-Touch Attribution (MTA)

Multi-Touch Attribution is a bottom-up, user-level measurement method that distributes credit for a conversion across the multiple ad touchpoints a customer interacted with on the way to purchase, rather than awarding 100% of credit to a single interaction (as last-click does). It is the user-level, tactical sibling of Media Mix Modeling (MMM) (top-down, aggregate) and Incrementality (experimental, causal) in the vault's measurement cluster — and the one most degraded by the privacy era. Per Web — Multi-Touch Attribution (MTA) 2026-06-27|Measured, MTA's mechanics "mostly held together" only roughly 2012–2018, when third-party cookies worked and people used one device.

Firewall: every claim below is what a source reports. See ../../CONTEXT.md Rule 1. This run's evidence base is heavily vendor-skewed — the loudest "MTA is dead" voices sell the replacements. Read accordingly.

Attribution models

Per Web — Multi-Touch Attribution (MTA) 2026-06-27|Improvado, the common credit-allocation rules:

Model	How credit is split
First-touch	100% to the first known touchpoint
Last-touch	100% to the final touchpoint before conversion
Linear	Evenly across all touchpoints (e.g. 20% each in a 5-touch path); "flattens the nuances"
Time-decay	Exponential weighting toward touchpoints nearer conversion; half-life (7/14/30d) is "arbitrary and dramatically changes credit allocation"
Position-based (U-shaped)	40% first touch, 40% last touch, 20% split across the middle
W-shaped	30% each to first touch, lead-creation/MQL milestone, and conversion; 10% across the rest — built for B2B funnels
Data-driven / algorithmic	ML assigns credit by each touchpoint's marginal contribution (journeys with vs without a channel), often via Shapley values or Markov chains

Per Improvado, data-driven models need ~2,000–3,000 conversions/month for statistical validity ("not viable for most businesses") and their black-box weighting reduces stakeholder trust.

Why MTA is in decline

Per Web — Multi-Touch Attribution (MTA) 2026-06-27|both vendor sources, privacy changes (Apple iOS 14.5 / ATT, third-party cookie deprecation) and walled gardens have eroded the user-level signal MTA depends on.

Benchmarks (as-of 2026-06-27)

• MTA coverage has shrunk to 30–60% of its 2020 signal — corroborated independently by Improvado AND Measured.
• Apple ATT opt-in rates stabilized at 15–25% globally as of Q1 2026 (Improvado, single source).
• Some advertisers reported losing visibility into 40–60% of iOS conversions after ATT (Measured).

All figures are vendor-sourced; no independent analyst benchmark was retrieved this run.

[!unverified] Snippet-only Gartner 2025 UK Digital Marketing Survey figure — only 24% of UK B2B orgs use MTA. Original Gartner source not located.

Per Measured, walled gardens (Google, Meta, Amazon, TikTok) do not share raw user-level data with third-party attribution vendors, leaving structural blind spots where the largest platforms sit.

Methodological criticism

Per Web — Multi-Touch Attribution (MTA) 2026-06-27|Measured (a vendor that sells the alternatives):

• MTA measures correlation, not causation — it cannot run the counterfactual (the world where the ad never ran), so it cannot say what an ad actually caused. "A more sophisticated version of a flawed idea is still a flawed idea."
• MTA over-credits retargeting and branded search (touchpoints cluster near conversion) and under-credits prospecting/upper-funnel — so cutting prospecting quietly depletes the retargeting pool and creates an unexplained growth ceiling.
• Per Improvado, attribution-window choice (7-day vs 90-day) can shift channel credit by 20+ percentage points — in one worked example paid social gets 0% credit under a 7-day window vs 20% under 90 days.

[!unverified] Improvado cites unnamed "industry surveys" that 30–40% of B2B buyer touchpoints occur in untracked channels (analyst calls, peer referrals, review sites, LinkedIn DMs) — credited at zero by MTA.

MTA vs MMM vs incrementality (triangulation)

Per Web — Multi-Touch Attribution (MTA) 2026-06-27|Measured, the primary replacements are Media Mix Modeling (MMM) (aggregate spend vs outcomes, no cookies/pixels) and Incrementality testing (controlled experiments measuring causal lift). Measured concedes MTA is "not completely" useless: at the campaign-management level it still gives directional signals for creative testing and tactical optimization within a single platform, but breaks down as the basis for cross-channel budget allocation.

[!unverified] Snippet-only (House of Martech, deep fetch failed): no single method suffices in 2026 — MMM for quarterly/annual budgeting incl. offline; incrementality as the causal "gold standard" before scaling; MTA for daily campaign-level optimization within already-validated digital channels. Triangulation via Bayesian calibration (incrementality results feed MMM as priors).

A practitioner validation cited by Improvado: for any channel receiving >25% of attribution credit, run a 2-week holdout; if conversions decline <10%, the model is over-crediting that channel for demand generated elsewhere. (This is essentially folding Incrementality back into MTA QA.)

Contradictions

Key terms

Term	Meaning (per sources)
Last-click / last-touch	100% of credit to the final interaction — the baseline MTA improves on
Data-driven attribution	ML credit allocation by marginal contribution; needs high conversion volume
Shapley values / Markov chains	Probabilistic methods underpinning data-driven MTA
Attribution window	Lookback period (e.g. 7d/90d) over which touchpoints count — strongly affects credit
Walled garden	Platform (Google/Meta/Amazon/TikTok) that withholds user-level data from third parties
Triangulation	Combining MTA + Media Mix Modeling (MMM) + Incrementality rather than trusting one

Frontier links

Multi-Touch Attribution (MTA) completes the run-103→106 measurement triangle. Still-dangling siblings: ROAS (referenced everywhere since run 103, still no page) · Data-Driven Attribution (DDA) · Shapley Value Attribution · Markov Chain Attribution · Apple App Tracking Transparency (ATT) · Walled Gardens · GA4 Attribution · Showrooming (carried from run 102) · Channel Conflict (carried runs 100–106).