Attribution & Measurement

Multi-Touch Attribution (MTA)

Also known asMTAMulti-Touch ModelingAttribution ModelingFractional AttributionFractional Credit

An attribution model that distributes install credit across multiple touchpoints in the user journey — not just the last click.

Key takeaways

  1. 01Last-click attribution gives 100% credit to the final ad clicked. MTA distributes credit across multiple touches.
  2. 02Common MTA models: linear (equal split), time-decay (more recent = more credit), position-based (40/40/20 to first / last / middle), data-driven (ML-based weight).
  3. 03MTA is harder on mobile post-ATT — SKAN is essentially single-touch; cross-channel multi-touch needs device-level signal that's now restricted.
  4. 04Incrementality testing (holdout experiments) is the more rigorous alternative when MTA signal isn't available.

Most mobile attribution defaults to last-click: the final ad the user clicked before install gets 100% credit. Simple, easy to implement, and the default in every MMP. But the user journey is rarely single-touch — a user might see a TikTok ad on day 1, a YouTube ad on day 3, click a Meta retargeting ad on day 5, then search the brand and install. Last-click gives Meta all the credit; the upstream creators (TikTok, YouTube) get none, even though they generated awareness.

Multi-touch attribution (MTA) distributes install credit across multiple touchpoints. Common MTA models:

  • Linear: equal credit to every touchpoint. 5 touches → 20% each.
  • Time-decay: more recent touches get more credit. The last click might get 50%, the previous 25%, the one before 12.5%, etc.
  • Position-based (U-shaped): first and last touches get the most credit (typically 40% each); middle touches split the remaining 20%.
  • Data-driven: machine-learning model that assigns weights based on observed correlation with conversion. Generally most accurate but requires significant data and platform support.

MTA is significantly harder on mobile post-ATT. The problem: cross-channel multi-touch requires you to know that User A saw a TikTok ad AND a Meta ad AND a YouTube ad — which requires a stable identifier across networks. Post-ATT, that identifier isn't reliably available on iOS for the 60-80% of users who don't opt in. Each network (Meta, TikTok) has its own attribution view but they can't be unified at the user level.

Workarounds in 2026

Bottom line in 2026: pure multi-touch attribution at the user-journey level has degraded on iOS. Most sophisticated mobile UA teams now run a combination: last-click attribution as the operational source of truth, plus incrementality testing for big strategic decisions, plus MMM for top-down channel allocation. MTA models persist mostly inside walled gardens (Meta, TikTok internal MTA) rather than across them.

Quick answers

What is multi-touch attribution?

Multi-touch attribution (MTA) distributes install credit across multiple touchpoints in the user journey — not just the last click. Common models: linear (equal weight), time-decay (recent touches weighted higher), position-based (40/40/20 to first / last / middle), data-driven (ML-learned weights). Contrasts with last-click attribution, which gives 100% credit to the final touchpoint.

Why is multi-touch attribution harder on mobile?

Multi-touch attribution at the user-journey level requires a stable identifier across networks (to know the same user saw ads from Meta AND TikTok AND YouTube). Post-iOS-14.5 ATT, that cross-network identifier isn't reliably available for 60-80% of iOS users. Within-network MTA still works (Meta can do internal multi-touch within its own properties), but cross-network MTA is degraded.

What is the difference between MTA and incrementality testing?

**MTA** attempts to assign credit across observed touchpoints — it's a retrospective allocation of credit. **Incrementality testing** measures actual causal impact: a control group sees no ads, a treatment group sees the campaign, the lift in conversions IS the campaign's incremental impact. More rigorous than MTA (it measures causation, not correlation) but requires holdout-group infrastructure and is slower / more expensive to run.

What is data-driven attribution?

Data-driven attribution is an ML-based MTA model that assigns weights to touchpoints based on observed correlation with conversion outcomes in your specific data. Generally more accurate than rule-based models (linear, position-based) but requires significant data and platform support. Google's Attribution 360 and Meta's data-driven model use this approach within their ecosystems.

Back to glossary