What is data-driven attribution in GA4?

Data-driven attribution is GA4's default way to split credit for a sale across the ad clicks and visits before it. It learns from your property's paths, including people who did not buy, and gives each touch a fraction. If nobody changed the setting, it is the model behind GA4's key event reports.

The usual answer adds that it is the smart model, so you can trust it. Half of that holds. It is smarter than handing everything to the last click. But it still shares out credit for sales GA4 recorded. It never switches one of your channels off to see what happens.

What one store's data shows

One store's anonymised GA4 export, 1 January 2024 to 21 August 2026. It holds shares of revenue only: no ad spend, no order counts.

The touched view shows what attribution looks like without fractions. It gives every channel the full value of each journey it touched, so the column adds up to 110.4% of revenue (Channels sheet). Data-driven attribution does the opposite. It splits each sale, and Google's help says the shares for one key event sum to 1.0.

Journeys with one touch hold 79.5% of revenue and took 0.5 days to buy (Journeys sheet). There is nothing to split in a journey of one. Data-driven, last click and first click all hand that money to the same channel. For most of this store's revenue, the model is a bystander.

The export lists 3,670 distinct path sequences (Journeys sheet). Of those, 3,656 have two or more touches, yet together they hold about a fifth of revenue (Journeys sheet). That slice is the only ground where data-driven attribution can move credit. Distinct paths are not purchases, so these counts say nothing about how many people bought.

Here is the part the export cannot do. It cannot rebuild GA4's data-driven numbers, and the reason is a short lesson in how the model works. Google's paths report only holds journeys that reached a key event. Data-driven attribution also learns from the paths of people who never bought. Those journeys sit in no paths export, this one included.

So the export shows where credit could move: the multi-touch journeys. It cannot show how GA4 moved it, or whether any touch caused a sale.

Why does the usual answer mislead?

The usual answer says data-driven attribution measures each channel's real contribution. Google's own description is more careful, and worth reading slowly.

The model compares converting and non-converting paths. It weighs factors such as time from the key event, device type, the number of ad interactions, their order and the type of creative. Then, in Google's words, it uses a counterfactual approach to judge which touchpoints are most likely to drive key events.

Google's help gives an illustration. Four ad exposures together lead to a 3% probability of a key event. Without the fourth, it drops to 2%. So the fourth exposure is credited with adding half again to the odds, and the model repeats that for every touch.

That is a sensible way to share credit. But look at what gets compared: paths with a touch against paths without it, among journeys GA4 recorded. A Meta ad seen and never clicked leaves no touch in GA4, so it sits in neither group.

Google also says its probability models are trained on data from randomized controlled trials. The help page describes those trials for Google ad exposures. It mentions none for your email, your Meta ads or your creator posts.

Two more details trip people up. Direct gets credit only when the whole path is direct visits; otherwise the model leaves it out. And a switch of reporting model rewrites history: Google says the change applies to past and future data. Last quarter's report can change without anyone touching a campaign.

What can data-driven attribution not tell you?

It cannot tell you what a channel caused. The model estimates which touches most likely helped, from patterns in recorded paths. Pausing a channel in some regions and comparing total sales measures what it adds. Those two answers can differ, and the second is the one your budget needs. Incrementality testing covers how that test works.

It cannot see what never became a touchpoint. A podcast mention, a shop shelf, a creator video nobody clicked: none of them reaches GA4 as a touch. Their sales land on whatever was clicked later, or on Direct.

It cannot explain itself. GA4 shows the credit each channel ends up with, not the weights behind it. Numbers can also shift after the fact: Google says conversions can be reattributed for up to 7 days after they happen.

And it cannot promise it had enough data. GA4's help says that, depending on data availability, the model may draw on aggregate data from data sharing settings. Google Ads recommends at least 200 conversions and 2,000 ad interactions in 30 days for its own data-driven model.

What to do this week

1. Confirm which model built your reports. In GA4, go to Admin, and under Data display click Events, then Attribution settings. Read the Reporting attribution model and the Channels that can receive credit. Pass: it says Data-driven, and your team knows it. Fail: nobody knew which model built last quarter's numbers.
2. Run it against last click, purchases only. Open Advertising, then Attribution, then Attribution models. Pick purchase in the key events menu at the top left, because by default every key event is lumped together. Pass: your biggest channels barely move in the % Change columns. Fail: one swings hard, so its credit depends on the model rather than on buyers.
3. List the touches the model cannot see. Write down every paid channel that people mostly watch rather than click: creator videos, podcasts, social video. Then search for each one in Attribution paths, under Advertising, then Attribution. Pass: each appears in paths that carry revenue. Fail: one never appears, so data-driven gives it nothing, and only a holdout can price it.

Check the homework. Your GA4 Attribution paths export already holds the evidence. Causality Engine reads that one file and shows what each channel caused next to what last-click gave it, in 1 to 2 minutes, for €99 once (excluding VAT), refundable within 30 days. Check the homework

Sources, 1 October 2026: Get started with attribution (Google Analytics Help); Select attribution settings (Google Analytics Help); Key events attribution paths report (Google Analytics Help); Key event attribution models report (Google Analytics Help); Default channel group (Google Analytics Help); About data-driven attribution (Google Ads Help).

Related answers

• All GA4 how-to questions
• How to set up data-driven attribution in GA4, step by step
• What does data-driven attribution mean for a beauty store?
• Is GA4's data-driven attribution a black box you can trust?
• Data-driven attribution, defined