A marketing attribution model is the rule that decides how the credit for a sale or conversion is distributed across the ads and channels a customer touched along the way. Last click is the most common attribution model and the most misleading one: it hands 100% of the credit to the final touch and zeroes out the channels that opened the journey. This guide covers how attribution models work, the limits of GA4's data-driven model, and which measurement to use for which decision.
What is an attribution model?
An attribution model is the rule or algorithm that splits conversion credit across the touchpoints in a converting customer journey (ad clicks, video views, emails...). Example: a customer watches your TikTok video, then three days later searches your brand on Google and buys. Last click gives 100% of the credit to search; multi-touch models spread it across the journey. Model choice is not a reporting detail: it is the primary input that decides which channel gets more budget.
Why does last click mislead?
Last click rewards the channel closest to the purchase, which is usually branded search and retargeting, because a customer who has already decided passes through them on the final step. Industry analyses show last click overstates paid search contribution by 40-65%, and that 60-80% of branded search clicks would have converted even without the ad. Discovery channels (social video, display, content) are systematically punished: they light the fuse, the closing channel collects the credit.
Which attribution models are left in GA4?
In late 2023, GA4 removed four models: first click, linear, time decay and position-based. Two options remain: data-driven attribution (DDA, the default) and last click variants. So the 'which model should I pick' question is largely settled inside GA4; the real question is whether the data-driven model actually works on your account.
How does data-driven attribution work, and where does it stop?
The data-driven model compares historical conversion paths and estimates each touchpoint's contribution to conversion probability (a Shapley-value approach with extra weight on recent touches). It is far fairer than last click, but it has a structural limit: it measures correlation, not causation. 'Saw the ad and bought' is not the same as 'bought because of the ad'. Compared with controlled experiments, click-based models (DDA included) can overstate impact by 2-10x, which is why incrementality testing remains the final referee for big budget decisions.
Why do platforms report different numbers?
Each platform counts its own conversions with its own window: Meta's default is 7-day click + 1-day view; Google looks at its own clicks; GA4 is session-based and sees all channels. The same purchase can land in both Meta's and Google's column, which is why the sum of platform reports exceeds real sales. That is not cheating, it is a definition difference; we broke it down in the GA4 vs. Google Ads conversion discrepancy article. The fix is not trusting one platform's number but building a shared roof: MER (total revenue ÷ total ad spend) is the simplest gauge of that roof.
Compare four platforms under one roof
Ads Sensor unifies Meta, Google, TikTok, Criteo and GA4 data in one panel; it computes the channel comparison and MER for you and watches for anomalies 24/7.
Which model should you use in practice?
There is no single correct model; what works is a ladder where each measurement does its own job. As of 2026, multi-touch attribution adoption has reached 47% and media mix modeling 26%; the direction is clear: layered measurement, not faith in one model.
- Read platform dashboards for direction: they are internally consistent and useful for creative and audience comparison; don't trust their absolute contribution claims.
- Do channel comparison in GA4's data-driven model: it is the only free roof that applies one rule to every channel (mind the 400-conversion threshold).
- Make MER your top gauge: total revenue ÷ total spend depends on no model; its monthly trend tells you the truth of the business.
- Track branded search separately: mixing branded and generic search in one campaign hides last-click inflation; split them and read branded with caution.
- Tie big decisions to experiments: let an incrementality test, not a model, decide whether to kill a channel or double its budget.
In short: an attribution model is a lens, not a truth machine; the last-click lens magnifies closing channels and erases discovery channels. Switch to data-driven, keep MER on top, tie big decisions to experiments. If you'd rather not run this comparison by hand, apply for the Ads Sensor beta and watch four platforms + GA4 in one panel with AI commentary.