Measuring what works

Data driven attribution: what it does, what it needs, and why you can't check it

Data driven attribution is an attribution model that splits the credit for a sale across the touches on the path using a rule learned from your own conversion data, rather than a fixed rule like last click or linear.

About the Google Ads connector

Google Analytics 4 and Google Ads both offer one.

Why it matters

The fixed models each pick a position on the path to reward. Data driven attribution sets out to reward the touches that made a difference, by looking at which touches tend to appear on paths that convert and which don't. If it works, channels that quietly make later touches more likely get credit for it. If it doesn't, you have handed the decision to a split you can't inspect.

How it works

The principle is comparison. Take all the paths that include a given touch and all the otherwise similar paths that don't, and compare how often each set converts. A touch whose presence goes with a higher conversion rate gets more credit; one whose presence changes nothing gets little.

Here is an illustration of the idea with made-up numbers. It shows the principle and nothing about Google's actual weighting, which is not published in a form you can reproduce.

Adding the email touch goes with 15 more orders per 1,000 paths. Under last click, email gets nothing on either path. A comparison-based model sees that paths with the email convert half as often again and gives email a share of the credit in proportion. Across thousands of such comparisons, every touch ends up with a share that reflects how much the conversion rate moved when it was present.

In Google Analytics 4 the model is a property-level setting, the reporting attribution model under Admin, then Data display, then Attribution settings, and data driven is the default; changing it changes the past as well as the future, because the reports are recalculated (Google Analytics Help, Attribution). In Google Ads it is chosen per conversion action, and it changes the conversions figure that automated bidding optimises against (Google Ads Help, About data-driven attribution).

What it can't tell you

It needs volume. The comparison in the table only means something because there were a thousand paths of each kind. With ten paths of each, the difference is a fraction of an order, and the model has nothing to learn from. Google Ads used to require 300 conversions and 3,000 ad interactions in 30 days before the model was offered, and dropped the requirement in September 2021 when it made data driven the default for every account (Google, Ads and Commerce blog, 2021). Google Analytics 4 publishes no threshold. So nothing stops the model running on thin evidence, and the only way to see what it is doing is the Model comparison report under Advertising, then Attribution: where data driven and last click give the same numbers for a key event, the model has found nothing to redistribute.

You can't audit it. A last click report can be checked by hand, by finding the last touch and crediting it. A data driven split can't, because the comparisons it made and the weights it chose are inside Google's system. When the split changes between two months, you can't tell whether customer behaviour changed or the model did.

It is still attribution rather than incrementality. It compares paths that were observed, and a touch can raise the conversion rate of the paths it sits on without having caused anything, because people who click an email were more likely to buy in the first place. Only an incrementality test separates those.

When to use it

Use it when you clear the volume thresholds comfortably and you mostly need the model for bidding inside Google Ads, where the platform's own optimisation uses it anyway. For the report you take to a budget meeting, keep a fixed model beside it, so that a shift in the data driven split can be compared against something that doesn't move on its own.

If you don't have the volume, don't force it. Use last click as the baseline and position based or time decay as the second view, as the attribution models page describes.

Tracking it in a spreadsheet

The inputs are conversions by channel under the data driven model and the same figures under a fixed model for comparison. A scheduled Fresh Sheets report can keep Google Ads conversions current in a tab of their own, each conversion action counted under the model you chose for it. The fixed-model figures you compare against go beside them on a separate tab, because Fresh Sheets replaces the report's tab on every refresh.