What is marketing attribution?

What you will learn

By the end, you will understand:

Explain attribution models and lookback rules
Diagnose why platforms disagree
Separate attributed credit from incremental causal impact

Visual explainer

See the idea clearly.

A customer path crosses social, search, email, direct and offline touchpoints; last-click and data-driven models assign different report credit while an experiment estimates incremental effect separately.

Attribution is a rule for credit

A customer can encounter several interactions before a key event. Attribution decides which observed eligible touchpoints receive reporting credit. The answer changes with model, identity, channel scope and lookback window.

Credit is useful for comparison and bidding, but it is not proof that the credited touchpoint caused the action.

Common model ideas

Model	Meaning
Last click	Gives credit to the final eligible interaction under the system’s rules.
First click	Gives credit to the earliest observed eligible interaction; unavailable in current GA4 attribution reports.
Rule-based multi-touch	Splits credit using predetermined linear/time/position rules; several were removed from GA4 in November 2023.
Data-driven	Uses platform/model data to assign fractional credit based on estimated contribution within available observations.

Lookback and scope alter the answer

A 7-day click window and a 30-day click window can credit different conversions. View-through, engaged-view, direct traffic and cross-device treatment also vary by platform and report.

Document settings and comparison dates. Do not compare numbers from different attribution definitions as if identical.

Why platforms disagree

Different tags/events
Different identity/cross-device view
Click versus view inclusion
Different lookback windows
Different time zones
Different attribution model
Consent/modeling
Deduplication
Import delay
Channel self-reporting scope
Offline conversion upload

Modeled is not observed

Google Analytics says modeled key events estimate unobserved activity using aggregate patterns under eligibility/quality conditions. Reports can update after the original event.

Label observed versus modeled where the product exposes it and avoid claiming the model reconstructed a specific person’s hidden journey.

Incrementality asks a different question

Attribution

How should observed credit be distributed among measured interactions?

Incrementality

How many outcomes happened because of the marketing compared with a credible no-treatment/counterfactual condition?

Use several evidence layers

01
Validate events
02
Compare attribution models
03
Reconcile with actual outcomes
04
Study journeys qualitatively
05
Run controlled experiments where feasible
06
Assess profit/capacity
07
Make decision with uncertainty

Real-world example

Example: email receives last-click credit

Example

A customer learns through a podcast and organic search, then clicks a reminder email before purchase. Last-click assigns email the sale; that does not prove the purchase would not have occurred without email. A holdout test can better estimate incremental reminder effect.

Try this

Compare two attribution views

Select one conversion and compare last-click with the available data-driven report. Record window, channel scope, modeled data and differences. Write three offline or unobserved influences.