Multi-Touch Attribution (MTA) is the family of attribution models that distribute conversion credit across more than one touchpoint on the path to purchase, assigning each touchpoint a weight w_i where the weights sum to 1.0 across the converting path. Practically, MTA sits between platform-level last-click and budget-level marketing mix modeling: useful as a conceptual frame for how channels interact, less useful as a number to plan budget against once the post-ATT signal gap shows up in the input.
How credit gets split
Credit for a given touch is w_i × conversion_value; models differ only in how w_i is set. The category splits into two branches by how the weights are chosen. Last-click is the degenerate case where the terminal touch takes w = 1 and everything earlier takes zero.
Rules-based models declare weights up front and ignore the data:
- Linear — credit divides evenly across touches.
- Time-decay — recent touches weighted more on an exponential curve.
- Position-based (U-shape) — first and last touches anchored, the middle shares the remainder.
- W-shape — adds a third anchor at lead creation.
Algorithmic / data-driven models fit weights to observed paths:
- Shapley-value decomposition — credit per touch is its average marginal contribution to conversion probability across all orderings of the path.
- Markov-chain removal-effect — credit per channel is the drop in conversion probability when that channel is removed from the graph.
- ML-fitted variants — what ad platforms ship; Google calls its version data-driven attribution and runs it as the GA4 and Google Ads default (since 2023).
The branch matters because rules-based models will produce the same channel split on any path of the same length and shape; algorithmic ones will not.
What data MTA needs
The weighting math runs on cross-channel event-level path data joined per user across the trailing lookback window. That join requires identity resolution — deterministic stitching via login or hashed email where available, probabilistic stitching via device and IP heuristics where it is not — to link touches arriving across browsers, devices, and apps as the same person. Single-touch attribution does not need any of this; MTA does.
Why operators use it less now
That input requirement is also the failure point. iOS App Tracking Transparency, Safari ITP, ad-blocker share, and consent-mode opt-outs mean the path data the weighting sits on is now systematically missing a structural share of touches — the gap is large on mobile-app inventory and on iOS Safari, smaller on logged-in surfaces. The credit assignment stays mathematically clean; the empirical bias in what gets fed into it is large enough that operators stopped trusting the channel splits MTA produces. The same effect explains why GA4, Northbeam, Triple Whale, and Rockerbox report different channel splits on the same brand — each stitches identity differently, observes a different subset of paths, and applies different weights. That divergence is a tool-difference question, not a truth question.
When MTA is enough
What teams reach for when MTA’s observational ceiling shows up is causal rather than observational: incrementality tests via audience holdouts, geo-lift testing for channels that cannot hold out at the user level, and marketing mix modeling for a top-down read on the whole portfolio. Those answer a different question — incremental revenue, not credit allocation — but they are what reaches the budget meeting. Treat MTA as the conceptual bridge between platform last-click and budget-level MMM, not the budget number itself.