Data-Driven Attribution (DDA) is the model that runs by default for most conversion actions in Google Ads. Credit isn’t distributed according to a fixed rule — it’s distributed based on the account’s own historical data.
The model’s core logic is simple: it compares the paths taken by users who convert against those who don’t. Whichever touchpoints are found to increase conversion probability the most receive credit accordingly.
The Google Ads data-driven attribution documentation explains that the model works by comparing the engagement paths of converting and non-converting customers — including clicks and video interactions across Search, Shopping, YouTube, Display, and Demand Gen. The same source notes that at least 200 conversions and 2,000 ad interactions within 30 days is the recommended threshold for the model to work well.
How Does Data-Driven Attribution Work?
The model doesn’t apply a fixed percentage rule — it learns from each account’s own data.
An example makes the comparison logic clearer
In Google’s own example, users who click a “Bike tour New York” ad and then click a “Bike tour Brooklyn waterfront” ad show a higher likelihood of purchasing than users who click only the second ad. The model detects that difference and distributes credit accordingly.
Cross-channel data is evaluated together
Clicks and video interactions across Search, Shopping, YouTube, Display, and Demand Gen are all evaluated within the same model. This means a single channel is read as part of the whole rather than in isolation — and that’s exactly where the real difference shows up.
How Much Data Is Needed?
A certain volume of data is recommended for the model to be able to detect patterns.
The 200-conversion, 2,000-interaction threshold
Google recommends at least 200 conversions and 2,000 ad interactions within 30 days. The model still runs below this threshold, but pattern detection gets weaker.
At low volume, grow traffic first
If an account is below this threshold, the right first move is to focus on increasing traffic and conversion volume rather than questioning the model directly. A distribution generated from too little data can be closer to a random guess than a reliable pattern — overlooking that difference can lead to the wrong budget decision. For example, in an account that generates only 40 conversions a month, the model might assign different credit to different channels every three months; that fluctuation may not reflect a real change in behavior, but simply an insufficient sample.
Which Accounts Benefit Most?
The model’s contribution scales directly with the length of the purchase path.
It makes the biggest difference in long, multi-channel journeys
In B2B or high-value product categories where users touch multiple channels repeatedly, the data-driven model can paint a far more realistic picture than last-click — and that’s exactly where the real benefit shows up, in these complex journeys.
The difference is small in simple, single-touch journeys
In simple journeys where a user clicks a single ad and converts immediately, the difference between the two models can stay relatively small.
Pre-Launch Checklist
When evaluating data-driven attribution:
- Is the account close to the recommended data volume thresholds (200 conversions / 2,000 interactions)?
- Are channel-level budget decisions being made based on this model’s report?
- If the model is running on low data volume, how cautiously is the result being interpreted?
Common Mistake
The most common mistake is treating the distribution produced by a data-driven model on a low-volume account as absolute fact and shifting budget heavily based on it. A distribution built from little data should be read as an order of magnitude, not an exact ratio.
Summary
Data-driven attribution is a model that distributes credit based on the account’s own data by comparing the paths of converting and non-converting users. It requires a certain data volume to run reliably; below that volume, results should be read cautiously. In long, multi-channel purchase journeys, it presents a more realistic picture than the last-click model.