Google Analytics is sunsetting 4 attribution models. The old models will be replaced with data-driven attribution in the coming months. Here’s the timeline:
- May 2023: For Google Analytics 4 properties, linear, first click, time decay, and position-based models will not be available for new conversion actions.
- June 2023: For Google Ads accounts, linear, first click, time decay, and position-based models will not be available for new conversion actions.
- September 2023: Google will sunset the four attribution models in Google Analytics 4 and Google Ads.
So, what happens to conversion actions that use the sunsetting models? According to Google, any conversion action that uses a near-deprecated attribution model will be automatically converted to the data-driven attribution model.
Advertisers can use the current last-click attribution model, but each conversion action must be manually changed.
Anyone still using the deprecated models in Google Ads will be affected. Since each advertiser’s data-driven attribution formula is different and not visible, anything other than last-click will be much harder to track.
In this article, we will explain more about attribution models in analytics, and discuss the data-driven attribution promoted by Google. We will show the issues with data-driven attribution and explain why you should compare different attribution models.
Attribution models are rules or algorithms that decide how credit for conversions is assigned to touchpoints on conversion paths. A conversion is attributed to individual clicks, ads, and factors along the user’s conversion path. Attributions have two models:
- Rule-based attribution – Conversion credit is divided between the relevant touchpoints in rule-based models. Attribution can be single-touch or multi-touch.
- Data-driven attribution – The goal is determining how many conversions and visitor journeys contributed to a sale. Because of this, the resulting model can differ significantly from simple rule-based approaches.
Attribution helps us understand which touchpoints in the conversion path play a significant role. The point is for you to avoid guessing and be able to credit the proper sources.
Let’s see a sample conversion path for someone who converts on the website.
- A user searches Google for “attribution models in GA4” and clicks on your blog post.
- The user returns to the same blog the next day and navigates to your YouTube channel from that blog.
- After a few more days, the user visits your website directly and joins your newsletter because they realize they can learn better that way.
Three sources contributed to this conversion path: Google’s organic search, YouTube video, and direct traffic.
So, which one should get credit for conversion? It all started with organic search, right? Or maybe the YouTube video convinced the user to sign up? Could it be the direct source, since that’s when they decided? Why not all three because they all contributed? That’s where attribution models kick in.
Each model will have different results on the same conversion path. There can be multiple touchpoints, and the conversion order can differ for every user.
Different attribution models indicate how conversion credit is applied. The standard attribution models are:
- First click: All credit for the first interaction. A visitor enters your website through Google Ads, looks around, and leaves. Later, they come back via an organic Google search link and fill out a form – a conversion. That last visit via organic search won’t be counted at all. The first interaction, via Google Ads, gets 100% credit for the conversion.
- Last click: It gives all credit to the last click before conversion. This model, for instance, would give all credit to organic search if an organic search followed Google Ad clicks.
- Last-non-direct-click models: These remove direct visits from the equation. It attributes 100% of the conversion to the latest known indirect click, traffic source, or referral. This is a variant of the last-click model.
- Time decay: This gives credit based on the time between interactions. It credits each touchpoint, but the closer the touchpoint is to a final transaction or achievement of a goal, the more credit it receives.
- Position-based: This model credits particular steps on the conversion path, typically first and last. It attributes X% of credit to the first touch, Y% to the final touch, and Z% to all touchpoints.
- Linear: A model that assigns equal credit across each step of the conversion path. Success is equally attributed to each referrer and visit. So three touchpoints – say Google Ads, Google organic search, and a direct visit – would see each touchpoint getting one-third of the credit before a conversion.
- Data-driven: It uses data to determine attribution credit. The model shifts depending on the unique path.
Many marketers rely only on the last click. There’s nothing wrong with this model, but it ignores touchpoints that may play a significant role in a conversion. A model with an all-or-nothing approach, such as the first click, the last click, or the last non-direct click, has this flaw.
The strength of the other three models – linear, position-based, and time-decay – is that they assign credit to multiple touchpoints. This doesn’t mean they are better at modeling conversion funnels. That said, it is worth spreading credit beyond a single point of contact.
Multi-channel conversion attribution gives credit for conversions in a customer journey spanning multiple channels.
There is no one best universal attribution model. It may be sufficient to use one attribution model depending on the situation. However, every business is different, uses different touchpoints, and communicates differently.
You probably invest in dozens of channels – social media, search advertising, email, blogging, etc. The customer journey is fragmented, and moving from first touch to conversion rarely happens within one browsing session. The process spans many channels and touchpoints and doesn’t happen overnight. Multi-channel attribution analysis is complicated, but attribution reports will give you plenty of indications over time if you design ingenious experiments around them.
All attribution models have their pros and cons. So it’s not about picking the best model the first time. To succeed, you must take the rough output, make changes, observe the impact (usually over weeks or months), identify insights, and become more accurate over time.
Testing multiple models, including custom ones, is the best path to finding the most suitable attribution model.
Multi-channel attribution reports are helpful. The attribution report will let you analyze which channels people used before completing your website goal or buying your product. Unlike basic acquisition reports, this report shows various attribution models and conversion paths. You can use it to determine which touchpoints generate the most valuable traffic to your website. Knowing this, you can better allocate your marketing budget.
The model comparison tool, available in the report, lets you compare attribution models such as last-click, position-based, first-click, last-non-direct-click, time-decay, linear, and custom models.
All in all, comparing different attribution models can help you understand how your marketing efforts drive conversions.
Founder & chief content creator at Deepskydata
“Marketing attribution is one of the most complex topics in marketing analytics. So, investigate and test it. Don’t use it as the complete truth. Attribution is always a simple model of a complex world. But learning about the various touchpoints your customers need to go through before buying is a crucial insight. You should start by looking at the conversion paths to understand what 80% of your customers do. You should also find out what are the most typical combinations. Then compare the different models and see how conversion rates change for specific channels. Discuss the data with your marketing team and ask them what model makes the most sense to them. You would ask: Wait, what?! Makes the most sense? Yes, since attribution models are trying to model a complex world, the marketing team can give feedback on what model looks best based on their experience.”
GA4’s default attribution model is data-driven. The model uses machine learning algorithms to analyze your Google Analytics account data. It assigns credit to different touchpoints based on their actual impact on conversions.
GA4 deprecates non-last-click rules-based attribution models, including first-click, linear, position-based, and time decay in Google Ads and GA4. But cross-channel data-driven attribution, cross-channel last click, and ads‘ preferred last click will continue to be available.
If you use sunset attribution models, there is a workaround. Google Analytics 4 allows you to export event data into Google BigQuery. This will enable you to build your first-click, linear, time-decay, and other rules-based marketing attribution models using logic you own and can change under your organization’s control, not Google’s. But it requires additional time and resources.
It is worth noting that using attribution models also requires a lookback window.
Lookback windows, called attribution windows, help you to determine which ads resulted in conversions during the specified time frame. They also allow you to choose how far back you want to trace your users’ visits to your page or site. For example, if you use a lookback window of 30 days, GA will only consider user visits within 30 days before conversion.
By default, GA4 uses a 30-day lookback window for acquisition conversion events and a 90-day lookback window for other conversion events. So similarly, a 90-day lookback window means touchpoints can be qualified for a conversion credit for up to 90 days from the day they occurred. Universal Analytics has a six-month lookback window.
Due to this difference, there may be data discrepancies between Google Analytics 4 and Universal Analytics. For example, if a user clicks through to your website from a social link today, revisits it in four months, and converts, Universal Analytics would assign conversion credit to the social click. Google Analytics 4 gives conversion credit to direct. That might cause migration problems and data discrepancies.
You may also like: GA4 migration problems: What are the alternatives
Over the past few years, major ad platforms have gradually introduced data-driven attribution into conversion measurement. The motivation for this is to preserve performance tracking. Platforms must show their value when traditional tracking methods are lost due to third-party cookie deprecation.
Data-driven attribution offers some benefits over rules-based models. For example, it removes human bias from consumer touchpoint analysis by using Google AI. It helps to identify patterns that lead to conversions. And it automatically adjusts attribution weights as your marketing landscape changes.
It is, however, important to note that having a marketing attribution model that changes the weighting it applies to email marketing interactions from one year to another can result in useless comparison reporting. You will never know if the increase in conversions attributed to that channel was due to your increased spend or to Google’s change in attribution.
The data-driven attribution model is the most widely used conversion model in automated bidding, according to Google.
The four deprecated models comprise less than 3% of Google web conversions. In their view, removing less commonly used attribution models simplifies and consolidates analysis.
There is more to that – a data-driven attribution model may credit Google Ads campaigns more for conversion than other channels, so advertisers are cautious. If it makes measuring or justifying other paid media channels difficult, marketers could be induced into investing more of their marketing budget in Google Ads.
Let’s say a customer clicks on an email, searches for your brand, and watches a YouTube video. They then click on Facebook and go to the website to buy.
In this scenario, Google will only give credit to Google properties. As a result, your conversion tags or analytics platforms may over-credit a conversion or double-count a conversion.
Considering all of the above, it is no wonder that the news about sunsetting attribution models has generated a lot of discussion among marketers and advertisers. Removing attribution models takes away necessary insights for marketers that help them make more informed decisions. Moreover, many argue that data-driven attribution might be used for optimization, but comparing results using different models is what provides truly valuable insights.
Let’s discuss some other problems with data-driven attribution.
Attribution modeling relies on smaller data sets. When statistical models are built on a small data set, they’re less reliable, and the actual numbers can deviate significantly from the modeled data. More conversion reporting will be based on modeling untracked data. Despite their best efforts, marketers make gut decisions due to unreliable tooling.
You can’t see how attribution is modeled
The model considers factors like device types, how many ad interactions users had, ad exposure order, the type of creative assets, and the time from conversion.
It uses a counterfactual/imprecise approach in terms of what could have happened versus what actually happened to determine which touchpoints are most likely to drive conversions. It then gives credit based on this conversion probability. Google doesn’t reveal how data-driven attribution assigns credit to each channel.
All of this makes it quite challenging to understand what’s under the hood of its algorithm compared to other models. Marketers will have to trust the information that Google Ads gives them without seeing the inside of the process. It is necessary to accept a certain level of uncertainty when using data-driven attribution.
As for Google’s data-driven attribution model, the main downside is that some types of conversion actions need at least 300 conversions and 3,000 ad interactions in supported networks within 30 days to be eligible. You won’t be able to continue using data-driven attribution for these conversion actions if your data drops below 2,000 ad interactions in supported networks or 200 conversions for the conversion action within 30 days.
Unfortunately, this is going to be too much for many businesses.
Data-driven attribution is a black box. You can’t see how attribution is modeled and how data-driven attribution assigns credit to each channel. Applying machine learning is not bad, but the lack of transparency is a concerning trend. Many marketers use different attribution models and compare them regularly to find insight into potential issues and to form better assumptions about the relative importance of various channels.
Digital analytics consultant
“The main issue with data-driven attribution (DDA) provided by Google Analytics is that when using DDA, we trust our advertising budget and attribution to the same company without seeing what happens in the black box. The more our analytics data is based on modeling (instead of measuring), the harder it becomes to understand the models, methods, and limitations. DDA is helpful in some ways. It simplifies the analysis and provides marketers with results they wouldn’t easily be able to derive from their limited data. But they shouldn’t blindly trust the data, especially when companies running big advertising platforms provide it.”
With Google’s decision to remove attribution models, marketers must explore alternative solutions to ensure data-driven decisions.
This change will impact advertisers who still use near-sunsetted models. Models other than the last click will be challenging to track as the data-driven formula is account-specific and unclear. We don’t see how it works, what rules it follows, or why it takes specific actions.
One solution is to turn to platforms like Piwik PRO. This platform offers advanced multi-channel attribution tools to help you analyze which channel or touchpoint should get credit for conversions. Besides predefined attribution models, you can get custom ones to fit your business needs. And the comparison tool lets you compare different attribution models.
Last but not least, precise attribution requires collecting personal data. The emergence of increasingly strict regulations such as CCPA and GDPR makes this a very delicate operation. So a built-in consent manager is essential to your attribution toolbox. It will only get more critical as data privacy laws tighten worldwide.
See product comparisons that will help you choose the right web analytics software. We present the essential features of all products, explaining in detail their significance from the legal, technical, and practical perspectives: