📈 Analytics & Metrics

Attribution Model

attribution model · conversion attribution model · last click model · first click model · data-driven attribution · atribution model

Definition

An attribution model is the algorithm for distributing the value of a conversion across the marketing channels involved in the user's path to a purchase.

Why an attribution model is needed

A user rarely buys after the first touch. A typical e-commerce path: saw an article in search results → came through a social feed → opened an email → tapped a browser notification → bought. Which channel “earned” the conversion?

An attribution model is the rule for that distribution. The choice determines how a marketing team judges channel ROI and where it sends the budget.

The main models

Model	Logic	When to apply it
Last click	100% of the value to the final touch	Short purchase cycle, brand search as the main channel
First click	100% to the first touch	Focus on awareness, assessing acquisition channels
Linear	Split evenly across all touches	Long purchase cycle, no obviously key channel
Time decay	More value to the later touches	Purchase cycle of 2–4 weeks, conversion depends on a final nudge
Data-driven	Machine learning weighting on real data	Sufficient conversion volume (400+ a month), GA4

Data-driven attribution

The GA4 model uses machine learning to estimate the real contribution of each touchpoint. The algorithm compares converting and non-converting paths and computes a probabilistic contribution for each channel.

Important: data-driven attribution requires a minimum data volume (GA4 recommends at least 400 conversions and 4,000 events a month at property level). With insufficient data GA4 falls back to last click automatically.

Attribution in personalization

Recommender systems and onsite personalization need their own attribution logic, separate from marketing channels. Here attribution works at the level of widgets and experiments:

User → clicked a recommended product (T=0)
→ returned to the site (T+3 days)
→ bought the product (T+5 days)
→ the purchase falls inside the attribution window (7 days) → credited to the widget

An A/B test of recommendation strategies compares attributed revenue across variations inside the defined window.

Common mistakes

Comparing channels under different models. If one report counts on last click and another on linear, the numbers are not comparable. Fix a single model.
Ignoring cross-device paths. A user who started on mobile and bought on desktop is attributed incorrectly without identity resolution.
Too short an attribution window. In categories with a long decision cycle — electronics, furniture — the standard 7-day window understates the efficiency of content channels.

Frequently asked questions

Which attribution model should e-commerce choose?

For mature teams with enough data volume, data-driven attribution in GA4. To start with, linear or time decay: both reflect the role of upper-funnel channels more honestly than last click. Keep last click only if your main channel is direct returns or brand search.

Why is last click the most common model and also a poor one?

Last click is simple to implement and gives an unambiguous answer. But it systematically undervalues upper-funnel channels — content marketing, SEO, display advertising — which create the initial interest. That leads to underinvestment in awareness and a budget skew towards closing channels.

What changed in attribution with the move away from Universal Analytics?

GA4 uses data-driven attribution by default instead of last click and counts conversions across the cross-channel path. The default attribution window is 30 days for clicks and 1 day for views. Data is recalculated retrospectively as delayed conversions arrive.

How does attribution work for recommendation widgets?

The contribution of recommendations is assessed with an attribution window, usually 7–14 days after the click. If a user clicked a recommended product and bought it inside the window, the purchase is attributed to the widget. An A/B test then compares recommendation strategies on attributed revenue.