1.8K members online now
1.8K members online now
Welcome to the Official Google AdWords Community
Find optimization tips and how-to guides and videos by the Google team, AdWords Experts and other industry experts
Top Contributor

The Holy Grail of Attribution – Data-Driven Attribution

This is part of our "5x5" series, in which we share five tips per week over the course of five weeks. We will share must-have tips about AdWords Editor, account optimization, ad-writing best practices, and much more. Make sure to explore the full series.

This week, we’ve explored what attribution is and how you can use it to improve the performance of your AdWords campaigns. We also took a look at attribution outside AdWords and how it can be used to improve your overall digital marketing efforts. Then, yesterday, we started to peel away the layers of the standard rules-based attribution models that Google provides, for free, to all users of Google Analytics.





Each attribution model that we discussed has its pros and cons, and is ultimately just another way to look at the same data. It can sometimes feel that attributing value based on how you want to look at it is somewhat arbitrary. To a certain extent, it is. The rules-based attribution models that we explored in the previous article in this series are all essentially just ways to look at the same conversion cycle and give weight based on what you know about your business and customers. If you get really creative, you can use a position-based model to assign credit to each touchpoint as a percentage of what you believe, or know, that channel contributed. However, it was still somewhat arbitrary.





Normally, my series would end here. However, I’m delighted to announce that as of last week, there’s a new kid on the block. One that solves all the problems of arbitrary appropriation in the default models and answers the question “which model is the right model for my business?” once and for all. Ladies and gentlemen, I give you the holy grail of attribution models, the biggest deal (in my opinion) in measurement since the tracking pixel, DATA-DRIVEN ATTRIBUTION.



Data-Driven Attribution


Data-Driven Attribution works just like the position-based model I mentioned in the previous article. However, with a big difference: Google examines the entire journey a user of your website or business took, and whether they converted or not. It then uses sophisticated machine learning methodologies to AUTOMATICALLY ASSIGN a value to the channel for you.






This data-driven approach across all channels (above) unfortunately is still only available for Google Analytics premium/Analytics 360 customers. I can tell you first hand that this feature alone is worth the steep cost of the upgrade.


But I’m not here to tease you about what features those with access can enjoy, I’m here to tell you that as of last week, Google released, in beta, this data-driven attribution feature inside AdWords for no additional fee. If we look back to the second article in this series, we can see how much of a role attribution plays in AdWords alone. Now, it’s available to everyone (almost) to benefit from.


Data Driven Attribution in AdWords


Attribution Model Dropdown - AdWords.png


If you were following along during the AdWords walkthrough, maybe you noticed it last week. If not, you can now access the data-driven attribution model and compare it with your other models, right inside AdWords itself. Just click on the “Tools” menu and then “Attribution.” Then select "Attribution Models" in the left-hand menu.


If you can't see this option available to you right now, don't fret. It's coming and it's going to be wonderful. Use the anticipation time to brush up on the previous models that we discussed so you understand them fully and make sure you're tracking all your conversions, no matter how small. Because with coupling this new way of attributing data with the enhanced ecommerce tracking features of GA and the new User Explorer in Google Analytics, this is going to be the biggest game changer for anyone with an AdWords account since quality score became a thing. It's an exciting time to be in the digital marketing space!


user explorer.jpgThe New User Explorer Feature in Google Analytics


How Does This Benefit You?


As an AdWords advertiser, you now have an extremely powerful tool at your disposal. You can now see, accurately, how each of your campaigns, keywords and ads contributed to your bottom line. Before, we could only see it at an aggregate level and make guesses as to how to split it out. Now we can see exactly how much value each has


"Data-driven attribution gives credit for conversions based on how people search for your business and decide to become your customers. It uses data from your account to determine which ads, keywords, and campaigns have the greatest impact on your business goals. You can use data-driven attribution for website and Google Analytics conversions from Search Network campaigns. This feature is currently in beta."


Google/The Model/AdWords takes a look at all your AdWords traffic. It then compares the clicks through your site that your customers took and compares that click path data to your site visitors that did NOT convert. It can then identify certain paths that have a higher likelihood of converting. The model then gives higher credit to those paths. Now, not only can you see how each campaign, ad group, keyword and ad contributed to a conversion, you can now start bidding confidently against that data. So that keyword with zero conversions that you had “a feeling” was the keyword that all your customers originally found you using, now you can know for sure. Now there’s one attribution model to rule them all, one that removes the need to guess or arbitrarily assign a value to each step in your funnel. It might even now settle some arguments in silo'd digital marketing agencies as to who provides the most value! 


AdWords Data-Driven Attribution Limits

Like the other models, the data driven attribution model is only available for search ads for now and of course, you need to have conversion tracking enabled or importing from GA. It’s also in beta, so you’ll need to take the weighting it spits out at you with a grain of salt for now. Google really refined its expertise in this area since it rolled it out to GA Premium/360 customer back in 2013. Now it’s available in AdWords and while it’s still in beta, and from personal experience, it’s quite accurate. Be sure to test the model and of course don’t be afraid to compare it to the models we explored previously to see if you knew your business and your customers as well as you thought you did. You might be surprised. 


Join me tomorrow for the final stop on our whirlwind tour of attribution. We’ll discuss attribution from a management perspective and how we can explain and demonstrate the importance of attribution to stakeholders who still think that the last click…is the only click that mattered.

about Dave Davis

Skydiving and travel obsessed. Director of digital marketing agency Redfly, based in Dublin, Ireland. An marketing agency that specializes in AdWords management, Google Analytics and Google Tag Manager.

Momchil N
May 2016
How about explaining the actual model behind this data driven approach. Are we supposed to religiously trust it? I see a worrying trends in marketing - casually mention "machine learning" and no further explanations are needed.
Dave_Davis Top Contributor
May 2016

Hi @Momchil N

A few of things.

1) This is a beginner series. I'm not going to get into the complexities of machine learning on a post introducing an attribution model. 

2) I don't work for Google so I don't know what the exact methodology is. 

3) If you don't want to trust "religiously" what data you're being given, collect it yourself. 

4) Nobody is asking you to "religiously" trust anyting. If you don't want to use it, don't. If you're not benchmarking or testing against a "control model" then you're not doing it right. But then again, if you're attributing based on the last click, you not doing it right either. 


There's a certain level of trust you must place in your analytics application anyway. I could just as easily say "How am I supposed to believe the session data in Google Analytics? Am i supposed to blindly trust the numbers? I see a worrying trend in marketing where people use third party software to get insight. They really should be counting the visitors themselves on an abacus". If you don't trust Google, that' fine. Nobody is asking you to. But if you have a better way to do data driven attribution in a spreadsheet at scale, I'd love to hear it. Google has the data to compare those who converted vs those who didn't. If you want to aggregate that data yourself, more power to you. 


So far, the bidding strategies and overall marketing strategies we've put in place based on this data have resulted in positive returns. If the machine learning methodologies used are conspiring against me, I don't care because the data, even if it's a placebo, is generating more revenue for our clients. 



Momchil N
May 2016
This post was featured by Adword's official twitter account, so I jumped to the conclusion that you are a product person. My bad. > I could just as easily say "How am I supposed to believe the session data in Google Analytics? There is plenty of documentation which describes the technicalities of data collection in google analytics. I can make myself familiar with this documentation and keep it in mind when I make my analysis. >I don't care because the data, even if it's a placebo, is generating more revenue for our clients. This is an absolutely ridiculous statement.
Dave_Davis Top Contributor
May 2016 - last edited May 2016

I should have stated, EVEN IF YOU THINK it's a placebo Smiley Happy

As for the technicalities, you're still trusting Google to collect it. 

I agree, the blanket term of "machine learning" is overused. But would you prefer I say "secret sauce"? Why throw algorithms 101 and 201 into a beginner topic for marketers, not computer scientists? 


You're asking for documentation that you know doesn't exist. You asked on the GA blog post and if Google are willing to share it, I guess they'll share it there. This series is supposed to be an introduction to attribution. Not sure if your issue is with the lack of visibility into the methodology or if you're just being condescending. I think we've interacted on the API forum before and I get it, but this post isn't a whitepaper. I think it's pretty obvious if you read the first few lines of the post. If you want to pull and compile the data that AdWords uses to gather this data to reverse engineer it, work away. But for me, like with conversion optimizer and other automated bidding strategies (like competitor outrank share) where the exact algorithm is not published and there's no way to know if Google is acting maliciously, I'm happy to trust Google as long as the results I'm getting with it are better than before. If you want to argue for the sake of arguing on this, let's move to SO but again, this is an introduction. If you feel the need to QA everything, you're free to create your own post. I don't get paid for this nor do I benefit from this. I usually just end up defending myself for trying to introduce topics to beginners. In future, I'll make sure to add a disclaimer to the top of my posts.


I'd be happy to collaborate on a post with you for a different audience where we can throw everything into R and play around but I doubt we can do that until the API method is released for the data. All we have to work with is the MCF reporting API right now: https://developers.google.com/analytics/devguides/reporting/mcf/v3/#intro and this is the AdWords community so no dice yet. 

Chris VP O
November 2016

Hi Dave


Great series this!


I've been playing about with our own attribution models but this is very time consuming therefore we are considering changing our conversion settings to data driven.  The one problem with this is how to measure the results for a business where 1 month to the next is not the same nor is year on year. Just wondering if you have any tips on measuring the sucess of such a huge change?


Thanks in advance

Samuel K

Has anyone used DDA within a Drafts and Experiments test?


Keeping everything constant (bids, ads, modifiers, etc), we've seen a big bias in our past tests making the Experiment side perform poorly.