Attribution models compared: last-click, data-driven, and the rest

An attribution model is a rule for splitting credit for one conversion across the several marketing touchpoints that preceded it. Every model is a simplification, none of them is true, and the choice matters mostly because it changes which channels look successful and therefore where budget goes. GA4 now defaults to data-driven attribution, which is better than the rule-based alternatives and still cannot tell you what would have happened if you had spent nothing.

The models, and what each one flatters

Last click. All credit to the final touchpoint before conversion. Simple, and systematically flatters bottom-funnel channels — branded search, retargeting, email — while making everything that created demand look worthless.

First click. All credit to the first touchpoint. Flatters awareness channels and ignores everything that closed the sale. Mostly interesting as a mirror to last click.

Linear. Equal credit to every touchpoint. Nobody believes all touches are equal, but it is transparent and hard to game.

Time decay. More credit to touchpoints closer to conversion. A reasonable compromise for short consideration cycles; still a guess about the decay rate.

Position-based. Typically 40% first, 40% last, 20% distributed between. Splits the difference between the two extremes by assertion rather than evidence.

Data-driven. GA4's default. Uses observed patterns — comparing paths that converted with paths that did not — to allocate credit. Genuinely better than rules, because it is derived from your data rather than asserted, and genuinely opaque, because you cannot inspect why it credited what it did.

Model	Flatters	Useful when
Last click	Branded search, retargeting	You need a simple, stable baseline
First click	Awareness, top-funnel	Diagnosing where demand originates
Linear	Nothing in particular	Transparency matters more than accuracy
Time decay	Recent touches	Short consideration cycles
Position-based	First and last	You want a defensible compromise
Data-driven	Whatever actually correlates	You have enough volume for it to engage

Why the model changes the answer

Take one customer: they see a display ad, search a generic term and click a paid result, return via organic search two days later, and finally convert after clicking an email.

• Last click credits email entirely. Display, paid search, and organic get nothing.
• First click credits display entirely.
• Linear gives each of the four a quarter.
• Data-driven allocates based on how much each touch type moves conversion probability across your whole dataset.

Same customer, same revenue, four completely different stories about which channel earned the budget. This is why "which model should we use" is really a budgeting conversation wearing an analytics costume.

What data-driven attribution actually does

It compares converting and non-converting paths to estimate each touchpoint's marginal contribution. Where a channel appears disproportionately in paths that convert, it earns more credit.

Its real advantages: it adapts to your business rather than importing someone's assumption, and it handles multi-touch paths without an arbitrary rule.

Its real limitations, which matter:

• It needs volume. Below a certain conversion count it cannot model reliably.
• It is not explainable. You cannot audit why a channel got 23%.
• It only sees what GA4 observes. Blocked, unconsented, and offline touchpoints are invisible, so the model allocates within an incomplete picture.
• It is still correlational. Correlation with conversion is not causation of conversion.

The limitation every model shares

No attribution model can tell you what would have happened if you had not run the campaign. They allocate credit for conversions that occurred; they cannot identify conversions that would have occurred anyway.

This matters most for the channels that look best. Branded search and retargeting reliably score well under most models, and both largely reach people who were already going to buy. An attribution report showing retargeting at a 12x return is not evidence that retargeting created that revenue.

The tools that do answer the causal question:

• Geo holdouts — suppress a channel in matched regions, compare outcomes.
• On/off tests over meaningful periods.
• Media mix modelling, if you have the scale and history.

Use attribution for allocation between similar channels, and incrementality testing for decisions about whether a channel should exist. Attribution windows covers the other half of the configuration.

Practical guidance

1. Use data-driven if you have the volume. It is the best default and requires no argument about coefficients.
2. Keep last click as a stable reference. Not because it is right, but because it is simple, unchanging, and useful for spotting when something structural moved.
3. Never compare periods across a model change. Annotate the switch. A model change looks exactly like a performance change.
4. Do not expect platform numbers to reconcile. Google Ads and Meta each use their own models over their own data, and each will claim the same conversion. Why they never match.
5. Test incrementality for the big decisions. Anything above a threshold you care about deserves a real test rather than a model output.

FAQ

Which attribution model is most accurate?

None is accurate in a causal sense. Data-driven is the best-calibrated because it derives credit from your own data rather than from an asserted rule, but it still allocates credit rather than proving causation.

Should I use data-driven attribution in GA4?

Yes, if your property has enough conversion volume for it to engage. It is the default and generally outperforms rule-based models for budget allocation between channels.

Why do my attribution numbers change when I switch models?

Because the model is the rule for splitting credit. The conversions are identical; only the allocation changes. Never compare periods across a model switch without annotating it.

Does attribution tell me which channels to cut?

Only weakly. Attribution shows correlation with conversion, not incremental contribution. For cut-or-keep decisions, run a holdout test — the channels that look best under attribution are often the ones with the least incremental value.

Why does Google Ads report more conversions than GA4's Ads channel?

Because Ads credits its own click-based model while GA4 splits credit across all channels. Both are internally consistent and answer different questions.

Before comparing models, make sure the conversions are being recorded correctly: the free tracking audit checks tags, duplicates, and consent signals on any URL.