Data-driven attribution (DDA) assigns conversion credit with an algorithm rather than a fixed rule. In GA4 it compares the paths of users who converted with those who did not, and estimates how much each touchpoint changed the probability of conversion. Google made it the default model in GA4. It is more nuanced than first- or last-click, but it is a model with its own limits, and its inner workings are not fully published.
How it works, broadly
Google describes DDA as analyzing the touchpoints in converting and non-converting paths, then estimating each touchpoint's contribution using counterfactual comparisons: how likely was conversion with and without this interaction? Credit is then distributed in proportion to those estimates. The exact model and its training data are not public.
Limits to keep in mind
DDA only sees what GA4 tracks. Offline sales conversations, private shares and visits blocked by consent choices are invisible to it. It also learns correlations in paths, not causal effects, so it cannot tell you what would happen if you cut a channel entirely.
- Only tracked, consented touchpoints count
- Results can shift as the model retrains
- Low-volume properties give the model less to learn from
- Correlation in paths is not the same as incremental impact
Using it well
Compare DDA with last-click to see which channels it moves credit toward, and treat large shifts as hypotheses. Validate big budget decisions with incrementality tests, which measure causal lift directly.
Example
Credit moving toward content
Imagine Bellwether Loom, a project tracking tool, compares DDA with paid and organic last click. DDA gives organic search noticeably more credit for trial sign-ups, mostly taken from branded paid search. The team treats that as a signal worth testing and plans a geo holdout for brand ads before changing budgets.
Common mistakes
Treating DDA as ground truth
It is a model of observed paths. It is better than a fixed rule for many uses, but it is not an experiment.
Comparing DDA in GA4 with DDA in an ad platform
Each platform runs its own model on its own data. They will not match.
Questions
Not directly. GA4 shows the resulting credit by channel and path, but not the underlying model parameters.
