Data Driven Attribution in Google Analytics 4
What is the best Attribution Model?? Ah, the age-old question…
Attribution is a big hairy topic, and I’m not going to write an essay on picking the best attribution model, because the best will always depend on your individual business… BUT, I’m excited to say that Google Analytics 4 now has Data Driven Attribution (DDA) available for ALL customers – both free and 360 customers!
In Universal Analytics (aka GA3), DDA was only available to 360 customers, so this is a great addition to the free version of GA4! (Side note – DDA is available in the Attribution Beta in UA for free for everyone, but since this is outside of the main Universal Analytics interface, I’m not including that here)
Let’s start with a few definitions: What is Data Driven Attribution?
Data Driven Attribution (DDA)
DDA is an algorithmic model that takes into account each of the touchpoints observed for your website conversion events. It then does some modeling (GA secret sauce) to assign credit to each channel. In GA3, DDA looked at the last 4 touchpoints to do this modeling, in GA4 that number goes up significantly to 50+ touchpoints, ensuring that all of your marketing efforts are taken into account when assigning credit.
Here is an example the Google Analytics 4 support page gives to explain DDA:
How does DDA compare to the ‘regular’ attribution model in Google Analytics reports?
Great question. To answer it, first we need to define what the regular attribution model in GA is… a short detour from our discussion on DDA:
Last Non-Direct Click Attribution
The “Last non-Direct click” model is the default in all non-MCF reports in Universal Analytics (GA3) and the User and Traffic attribution reports in GA4 and cannot currently be changed in these default reports (except for event conversion reporting in GA4, more details in a later post).
“Last non-Direct click” means that Google Analytics will look at the UTMs (source/medium) of the incoming session, and if it is anything other than direct, classify it to that channel. If it is Direct, GA will check to see if there is a previous session attribution to a channel other than Direct, and if so, it will persist the previous UTMs.
In GA4, this is true for the Traffic Acquisition reports which report on attribution at a session level. User Acquisition reports, however, will report on the last click attribution for the User which does not update unless the user cookie is reset (in which case they would be seen as a new user).
For example, if a user came to www.ksdigital.co first via directly typing the URL, and then later comes back via a referral from another website (www.example.com / referral), the first session will be attributed to the channel of Direct (direct /none), and the second session will be attributed to Referral.
However, if the user first came to www.ksdigital.co via a referral from another website (www.example.com / referral), and then the next day comes back directly by typing the website URL (direct / none), both sessions will show attribution for Referral.
Last interaction, credit given to Direct:
Last non-Direct Click attribution, credit given to Referral:
The “Last non-Direct click” attribution model that is core to Google Analytics therefore leads to a large redistribution of traffic into other channels that may have actually come back via Direct in a “last interaction” model. You can actually see the impact of this on your business by comparing a “Last interaction” model to a “Last non-Direct Click” model in the Universal Analytics Model Comparison tool.
As you can see, there is a decrease in the number of conversions attributed to direct traffic when using the “Last non-Direct click” model, and increase across every other channel where this traffic has been redistributed to.
Breaking down “Last non-Direct Click”
Last non-direct click attribution gives credit for the long-term effects of marketing campaigns. Many organizations, particularly those with long buying cycles, like to see these long-term effects.
In addition, last non-direct click attribution decreases the amount of traffic that is labeled “Direct” in reports while increasing traffic to known sources of traffic. This can give you more actionable data when putting together your marketing strategy. It can be much easier to react to data about the performance of your email, social, or cpc campaigns than to the performance of direct traffic.
While last non-direct click attribution increases the amount of data you have on known traffic sources, this data can be misleading, hard to interpret, or raise questions that you just can’t answer. For example, did that user really click on an email link 4 months after it was sent, or are they just returning to the site directly? With last non-direct click attribution, we cannot answer that question.
By taking away credit from direct traffic and giving it to known sources of traffic, Google is subtly increasing the value of known sources of traffic.
Back to DDA vs Last Non-Direct Click
As you probably gathered by the lengthy description of the Last Non-Direct click model, there are pros and cons, but using that model makes it hard to know the channel of what is really driving a conversion. Data Driven Attribution aims to assign credit across multiple channels in a way that more accurately demonstrates the impact of each channel on the ultimate conversion.
Let me be clear – DDA is a black box. You know there are a bunch of inputs (interactions of your marketing efforts resulting in site visits), and then there is some modeling behind the scenes using Google’s Machine Learning, and the output is conversions attributed to various channels.
But the black box is likely better. Last interaction, last non-direct click, or any other model you choose is never going to give you an understanding of the impact of multiple touchpoints on a customer conversion. That is why DDA is likely going to be a better model to help you gauge impact of multiple channels.
At the very least, now every customer of Google Analytics who is using Google Analytics 4 gets to be the judge for themselves as to which model they prefer, because DDA is now available to everyone in GA4!
To start using this model, I’d recommend checking out the model comparison tool in GA4 where you can compare Last non-direct click to DDA. You can click the drop down arrow to choose which model you are interested in or change from the default. Now that DDA has been released in GA4, the first column defaults to Cross-channel last click and the second channel defaults to Cross-channel data-driven:
Please note: This report (and also the Conversion Paths report) will default to using all of your conversions when modeling. This doesn’t make a ton of sense if you have a wide range of conversion types, for example, first visit and purchase. So definitely look at which conversions are being used in the report you are looking at and narrow it to what makes sense. In the screenshot above, I narrowed to just looking at the “Purchase” event, since understanding Conversions for Purchases by channel is one of the most meaningful things to look at here.
You can also choose to use the DDA model in the Conversion Paths report: