The New Normal: Data Driven Attribution (DDA) in GA4

Yes. We all know. We all have seen the message that Google is changing the default attribution model from Last Non-Direct Click to Data Driven starting this week in GA4.

And people are losing their minds. And I am unsure why. Most people have not been using GA4 long enough, even if this would be a seismic shift, to have anyone question the impact of this change.

But a few things. I’ve done spot checks of many accounts – Advertising – Model Comparison Tool – Cross Channel – Last Non- Direct Click to Cross Channel – Data Driven. And do you know what I’ve found- So far no material difference. Google even states, ‘ Note: Depending on data availability, cross-channel last click and data-driven attribution models can yield the same results in certain situations.’

Given that the Data Driven Attribution model was launched in November of last year, it is likely that you will also not see much difference until there is enough time for there to be a statistically significant amount data accumulated.

So – if you were worried that you would be getting questions that you couldn’t answer, fear not. The change will likely not have a significant impact yet. BUT if you are an attribution enthusiast, continue on with me.

I think some of the freak-out nature of this model is caused by its name – Data Driven Attribution. But to be fair, hasn’t it always, no matter which model, been data driven. I’d sure like to think so.

Having been in the attribution game for a minute, okay more like over a decade, the dirty little secret is that ‘straight’ math (call It linear) and ‘static’ algorithms were (somewhat) better than last click. And it was maybe not all that much better than last non-direct click, except when you needed to consider view-thru and offline data.

But now, to be intentionally overly dramatic, on the precipice of a cookie-less world – led by the browsers or other privacy advocacy initiatives, data driven becomes very important. With the absence of identifiers like cookies and the knowledge that outdated and incorrect static models are not effective, the world needs a bit more machine learning and artificial intelligence.