Google is currently moving all Google Analytics 4 users to ‘cross-channel data-driven attribution’. But what does this mean? And is it really the best way to accurately predict the results your campaigns will deliver?
If you use Google Analytics 4, you’ll probably have received this email recently:
Hello Google Analytics user,
We’re reaching out because your Google Analytics 4 properties are now eligible for cross-channel data-driven attribution. We’ll be upgrading your properties’ attribution settings to the data-driven attribution model on or after January 26, 2022. This change will apply to all of your Google Analytics 4 properties for reporting and measurement purposes. It will not impact your Google Ads campaigns.
Data-driven attribution is a machine learning model that attributes credit for your conversions based on how people search for your business and interact with your ads.
Compared with the last click model, data-driven attribution looks at the entire journey that leads a user to convert and allocates credit to each step in that journey.
You can learn more about how data-driven attribution may benefit your marketing performance and how to customize your attribution settings in Google Analytics 4.
The Google Analytics Team
As this email explains, on 26 January 2022 Google is finally ditching outdated last-click attribution and moving its analytics users over to something called ‘cross-channel data-driven attribution’, which is built on an algorithm that leverages machine learning.
Google’s machine learning model isn’t new
Google’s machine learning model isn’t new. Google Ads were already offering this data-driven attribution model, however until now it was not accessible to all advertisers due to minimum data requirements and some limitations on types of conversion.
Now though, Google are making data-driven attribution available to all Google Analytics 4 users, which sounds pretty good, right? After all, accurate attribution is essential to making wise marketing decisions, as Google themselves state in their white paper The definitive guide to data-driven attribution: “Assumptions about how marketing activities impact customers lead to a misinformed marketing strategy.”
But when you look more deeply into what Google is actually offering, all doesn’t seem quite as positive. In fact, we don’t think Google have gone nearly far enough with their new attribution model. Here’s why.
Strangely, Google won’t share the accuracy of their model
As we’ve just mentioned, Google’s machine learning model has been in use for some time. So they know just how accurate (or not) it is. But they’re not saying.
This is odd and leads us to believe that it’s not as accurate as they’d like advertisers to think. And rather than giving them granular, global insights they can act on with confidence, going into abstract ‘machine learning’ means users will actually have less idea of the interpretation. All they will know is that some channels are more valuable than others… for some reason.
That’s because, as we explain here, machine learning alone is limited: as clever as the programmes we build are, machines just can’t think like a human. In order to drive the best possible ad performance, you need both machines and humans.
So while this announcement may sound exciting, it’s not actually going to offer much practical use to advertisers.
Why Total Media Attribution is a far better approach
So what’s the alternative to Google’s machine learning approach? For a while now we’ve been using our own Total Media Attribution (TMA) attribution model with our clients.
The aim of TMA is to understand how media spend affects sales, and to optimise the allocation of spend across media to achieve the optimal media mix for our clients. TMA can track where conversions come from both online and offline, across a variety of different channels. It can measure the success of current campaigns and, more importantly, it can simulate conversion returns on future campaigns.
Our TMA model works across everything from TV, radio, print and PR to digital – including Facebook ads, Google Ads, programmatic, DOOH, CTV and audio.
And unlike Google, we DO share how accurate TMA is. Our accuracy of prediction reaches statistical significance (over 95%) and aims to be around 99%.
We can predict exactly where your returns will come from, and how you should weight your spending across your entire marketing campaign. We can even factor in seasonal variations and geographic preferences. So if people in Germany prefer print, we’ll upweight that. And if people in the US like video, our predictions will factor that in.
We can show you what returns your campaigns will get with each balance of media – and what will happen if you do nothing at all in a particular channel.
TMA was based on Google’s own theory
The irony in this is that the Total Media Attribution model we have developed was inspired by a white paper published by Google in 2017.
In the paper, Google proposed “a media mix model with flexible functional forms to model the carryover and shape effects of advertising.” They continued: “The model is estimated using a Bayesian approach in order to make use of prior knowledge accumulated in previous or related media mix models. We illustrate how to calculate attribution metrics such as ROAS and mROAS from posterior samples on simulated data sets.”
But mysteriously, Google seemed to leave this exciting proposition as a theory, and never developed into a working model. So we took the challenge and brought it to life as Total Media Attribution – which goes much further and is far more complex, with more potential for advertisers, than Google’s initial idea.
Google aren’t going far enough – and are letting down advertisers
While we think it’s good news that Google is moving away from a last click attribution model (we’ve long criticised this approach), we don’t believe Google is going far enough.
In our opinion models developed by Google have the capability to offer far more accurate insights and predictions on attribution, but Google are choosing not to use them. We believe that their current machine learning approach falls very far short of what advertisers deserve and could easily have from Google, and that is a shame.
Want to know more about Total Media Attribution and how it can help you plan campaigns with superior performance? Get in touch and we’ll be happy to help.