Marketing Attribution Models: 6 Types Compared With Examples

Marketing attribution assigns credit for a conversion to recorded marketing interactions. An attribution model determines how that credit is divided. First-click rewards discovery, last-click rewards the final eligible interaction, and multi-touch models distribute credit across several interactions.

Credit is not the same as incremental impact. A channel can appear before a purchase without causing it. Use attribution to understand recorded journeys and support campaign decisions; use incrementality tests to estimate what happens when advertising changes.

Six attribution models compared

Consider this illustrative journey: a customer clicks a paid social ad, later arrives through organic search, clicks an email, and finally clicks a branded search ad before a $100 order. Assume all four clicks are observable and eligible within the chosen lookback window.

The time-decay row assumes interactions 21, 14, 7 and 0 days before the order and a seven-day half-life. The weights are 0.125, 0.25, 0.5 and 1, divided by their sum of 1.875. It is a transparent worked example, not a universal setting. DDA percentages cannot be calculated from this single journey alone.

First-click vs last-click attribution

First-click attribution assigns the entire $100 to paid social. It answers which recorded channel first introduced this converting customer. It ignores the later interactions and misses any earlier contact outside the measurement window.

Last-click attribution assigns the entire $100 to branded search. It answers which eligible click preceded the purchase. It may be useful for understanding conversion routes, but it cannot tell you whether branded search created demand or captured existing demand.

Neither is inherently the true contribution of a channel. Comparing them reveals how model choice changes reported performance. Read why last-click attribution is misleading for the budget implications.

First-touch and last-touch are broader terms: a touch may include an eligible view or another event, depending on the system. Do not treat an ad impression, an email open and a website click as interchangeable without documenting eligibility rules.

Linear attribution

Linear attribution divides credit equally. Four interactions receive $25 each in the example. This makes the calculation easy to reproduce and prevents one interaction from automatically taking everything.

Equal weight is still an assumption. Repeated low-value interactions can receive more total credit than one useful interaction. Decide whether your model counts individual events, sessions or unique channels before interpreting the result.

Time-decay attribution

Time-decay gives recent interactions more weight. A common mathematical form is:

Weight = 0.5 ^ (days before conversion / half-life in days)
Credit = interaction weight / sum of eligible weights × conversion value

A shorter half-life concentrates credit near the conversion. A longer half-life spreads it more evenly. Choosing a half-life is a modeling decision; recency does not by itself demonstrate influence.

Position-based attribution

Position-based attribution reserves credit for the beginning and end of a journey. The familiar U-shaped version assigns 40% to each end and distributes 20% across middle interactions.

This can express a team's reporting convention around discovery and conversion. It does not establish that those stages caused 80% of the result. Also define what happens for journeys with only one or two interactions.

Data-driven attribution

Data-driven attribution estimates weights from data rather than using a fixed allocation. Google's documentation describes models that use converting and non-converting paths, counterfactual modeling and experimental training signals. It is therefore too simplistic to describe every DDA implementation as only counting correlations. Source: Google's attribution methodology.

The practical limitation remains: a platform attribution report is not a substitute for your own channel-level experiment. Its coverage, eligible channels, identity resolution and assumptions determine which effects it can represent. It cannot report a complete journey that was never observed or otherwise modeled.

There is no universal requirement of 3,000 monthly conversions for Google's DDA. Google removed minimum data requirements when expanding availability. Availability and estimate stability are different questions: sparse data still calls for careful interpretation. Source: Google's DDA availability announcement.

Which attribution models are available in GA4?

As checked in September 2026, Google's documentation lists data-driven, paid and organic last click, and Google paid channels last click in Analytics attribution reports. First-click, linear, time-decay and position-based models were removed in November 2023. They remain useful concepts and can be implemented elsewhere; they are not current GA4 menu options. Source: Google Analytics attribution documentation.

Keep acquisition reporting separate from key-event attribution. A report grouped by first-user source or session source answers a different question from a report allocating key-event credit. Changing a reporting attribution model should not be described as rewriting every acquisition report. Source: Google Analytics attribution settings.

Single-touch, multi-touch and multi-channel attribution

Single-touch gives credit to one eligible interaction. Multi-touch attribution distributes it across multiple interactions. Multi-channel describes the channel coverage, not a specific weighting algorithm. A multi-channel report can still use last-click.

A tool that collects first-party events from several channels may improve deduplication and reporting consistency. It still has gaps caused by consent choices, browser restrictions, offline interactions and missing identifiers. There is no defensible universal percentage for how much of every company's journey is missing.

How to choose an attribution model

Start with the decision, then choose the reporting method:

For implementation, use the attribution setup and QA checklist. For larger budget decisions, combine the report with MMM and controlled tests. MMM also requires causal assumptions and validation; it is not automatically correct because it uses aggregate data.