Why Google is Emphasizing Data-Driven Attribution Modeling and How You Should Respond

May 10, 2023

What to Do After Google Analytics First-Click Attribution Models Go Away in June 2023

Starting in June 2023, in Google Analytics you will not be able to select first-click, linear, time-decay, or position-based attribution models for Google Ads conversion actions not already using one of these models.  

After that point, you will be able to select only data-driven attribution (DDA), last-click, and external attribution models for new Google Ads conversion actions. Any conversion actions still using the deprecated rules-based models will be automatically switched to DDA in September 2023. 

To help you navigate this transition, I’ll first explain why it is happening and then provide some practical next steps.

Google’s emphasis on DDA is in line with broad shifts away from 3rd-party data and towards AI in the advertising industry

Many advertisers have had a good ride with the soon-to-disappear models, but now it is over. 

With the exception of the last-click model – which will continue to be an option in Google Analytics and can provide accurate but limited insights – rules-based approaches may do more harm than good, specifically when it comes to helping you understand increasingly complex consumer journeys with less 3rd-party data available. 

Take the first-click model, for example. If you are like most advertisers, until now it has provided you with valuable insights on long-term campaign performance and played a key role in prospecting. As the number of consumer journey touchpoints continues to rise and the gap between first acquisition efforts and conversions widens, however, this rules-based model will provide an increasingly distorted view of the consumer journey.  

Google’s AI-powered DDA model, on the other hand, is built for the rapidly evolving digital landscape. It uses machine learning algorithms to: 

Therefore, I see Google’s deprecation of less sophisticated attribution models as simply another step the company is making to leverage AI in a move from session-based analytics to user-based analytics. A previous step, for example, was the company’s November 2021 launch of Performance Max Campaigns, which use proprietary ML algorithms to automatically adjust ad bids, placement, and targeting across Google networks based on performance data.

This shift towards AI is visible across the entire digital advertising industry. It’s a big story, and Google’s pivot to AI-powered attribution modeling is one of the latest chapters.

How should you handle attribution modeling now?

In my view, you have two options, depending on your modeling goals and in-house capabilities:

  1. If you want to maximize accuracy and have – or are willing to invest in – the necessary know-how and resources (in particular, your own data lake), then you can take control of your own attribution modeling and avoid relying on the DDA “black box.”    
  2. If you want the best accuracy available from someone else’s attribution model, switch to DDA now. You will not know exactly how the model works, but you can be confident that your results will be “pretty good” and certainly better than what you’d get with a rules-based model that is being phased out anyway. Note that DDA supposedly requires at least 500 successful or unsuccessful monthly conversion events to train its machine learning algorithms properly. Like all AI-powered models, the more data you feed it the more powerful it becomes.

In Google Analytics 4, making the change to DDA is quite straightforward. It is now the default attribution model in the interface, but it is worth double-checking that it is currently set for each of your GA4 properties and ensuring that your reporting setup is configured accordingly. If you have any questions about navigating this transition to DDA or about Google Analytics attribution in general, my team and I look forward to getting them answered. Feel free to reach out.  

Not using Google Analytics 4 yet? You should be. Standard Universal Analytics properties stop processing data on 1 July 2023. We can help you cover all the bases for a successful migration. You can find out what that will look like in this guide:

Read the full report