The Ecomm Analyst

Data driven attribution is a model built by the ad platform from your own conversion paths, which assigns fractional credit to each touchpoint based on patterns it finds in paths that converted versus paths that did not. Can you trust it? Partly. It is better than the alternatives inside a single platform, and it is not something you should treat as an independent read on performance, because the company producing the number also sells you the media.

That is the whole tension. The model is genuinely more sophisticated than last click, and it is also unauditable and self interested. Most of the confusion I see comes from picking one of those facts and ignoring the other.

What it actually does

Rule based models decide credit in advance. Last click gives everything to the final touch no matter what. Position based gives 40 percent to the first and last no matter what. The rule is fixed before the data arrives.

Data driven models work the other way. The platform looks at large numbers of paths, compares ones that ended in a purchase against ones that did not, and infers how much each touchpoint type moved the odds. If people who saw a video ad converted at a meaningfully higher rate than otherwise similar people who did not, the video ad earns credit. The weights come out of the data instead of out of a rulebook.

Google runs this as the default in Google Ads and GA4. Meta does its own version, layered on top of modeled conversions, which is a separate thing. Modeling and attribution get discussed as though they are the same, and they are not.

Modeled conversions are a different problem

Since iOS 14 and the broader signal loss that followed, a meaningful share of conversions the platforms report were never directly observed. They are estimated. The platform sees a subset of real conversions, knows roughly what fraction of users it can still track, and scales up.

So when Meta reports 400 purchases, some of those are recorded events and some are statistical fill. The platforms do not break out the split for you. Then the data driven attribution model runs on top of that mixed pile and assigns credit. You are looking at a model applied to partly modeled inputs, and there is no way to unwind it from your side.

Where it breaks down for DTC

Three things go wrong on stores in the one to twenty million range. The first is volume. Data driven models need enough conversion paths to find patterns. A store doing a few hundred orders a month is thin. The model still returns numbers, and those numbers move around a lot more than the underlying reality does.

The second is that the platform only sees its own touchpoints. Google’s data driven model has no idea a customer saw four Meta ads first. It is distributing credit across the touches inside its own walls and calling that a full picture of the journey. It is not a full picture. It is a Google shaped slice of one.

The third is that it still measures correlation. If your branded search ads sit in front of people who already decided to buy, a data driven model will notice that those paths convert at a high rate and reward branded search accordingly. Correlation with conversion is exactly what the model is designed to find, and it is not the same thing as causing the conversion. Only a holdout answers that.

How I actually use it

I leave data driven attribution on inside the ad platforms, because the bidding algorithms use those signals and turning it off to get a cleaner report makes the campaigns worse. I just do not treat the output as my source of truth for what to spend next month.

My reporting number comes from order level data on my side. I use ThoughtMetric, which sponsors this blog, for that, because it reads orders directly from Shopify and matches them against sessions and click identifiers rather than accepting whatever each platform claims. Northbeam takes a heavier modeling approach if you want something closer to the platform philosophy but independent of any single one of them. Either way the tallying should happen somewhere that has no stake in the answer.

Then I use the platform numbers for what they are good at, which is relative comparison inside that platform. Which creative, which audience, which campaign. Those comparisons carry the same bias in the same direction, so the ranking stays useful even when the absolute value is not. Attribution windows shift that ranking more than people expect, which I went through in choosing attribution windows without overthinking it.

Common questions

Is data driven attribution more accurate than last click?

More sophisticated, and more reflective of the full path inside that platform. Whether it is more accurate depends on data volume. At low order volumes it is often noisier than last click.

Can I turn off data driven attribution in Google Ads?

Google has consolidated toward it as the standard for conversion reporting and the options have narrowed over time. Check your current conversion action settings rather than relying on what was available a year ago.

Why do my platform numbers still exceed my Shopify revenue?

Because each platform models and attributes independently, so the same order gets counted by more than one of them. Adding platform reported revenue across channels always overstates the total.