Basic Attribution Models
Credit for conversions, like purchases, can be attributed in many ways. Some marketers have mistakenly believed that the more complex an attribution model, the more valuable. This is not always the case. There are numerous attribution models that are calculated using simple arithmetic that can be valuable for marketers.
The Default Attribution Models in GA4 Are:
- Cross-channel last click: attributes 100% of the conversion to the last non-direct channel that a user engaged with
- Cross-channel first click: attributes 100% of the conversion to the first channel that a user engaged with
- Cross-channel linear: attributes credit for the conversion evenly to all channels that the user engaged with on the path to conversion
- Cross-channel position-based: 40% of the credit is assigned to the first and last channels, while the last 20% is evenly split between the channels in between
- Cross-channel time decay: channels that were visited closer to a conversion receive more credit
- Ads-preferred last click: attributes 100% of the conversion to the last Google Ads channel that the customer engaged with
This diverse set of default attribution models is valuable for the modern marketer. There are many different ways to use each of these models and this variety will suit a vast majority of users. However, there is one more additional model that Google offers that is a bit more complex and, for many, even more powerful.
Advanced Attribution With DDA
Data-Driven Attribution utilizes statistical models and algorithms to customize an attribution model based on each GA4 property's unique data. Google provides documentation on this topic, but I would like to explain this at a simpler level and attempt to demystify what this model is actually doing.
Step 1. Create a Conversion Probability Model
The DDA model starts by attempting to predict the likelihood that a user will convert, given their history of interacting with channels. This process takes all of the paths that users have taken and whether those paths led to conversion or not, and then can train a model that outputs a probability or likelihood that a specific path leads to conversion. This is fairly straightforward machine learning, and there are a number of routes one could take to predict a probability such as this. The method used to make this prediction isn't as important as how we interpret the model.
Step 2. Interpret the Probability Model
Once this model is trained and performing satisfactorily, DDA interprets the model using an algorithm developed by Nobel Prize winner Lloyd Shapley that yields a set of values known as Shapley values.
How Do Shapley Values Work?
It's fairly easy to find the precise equation for generating Shapley values with a quick Google search. However, I have found that, for those who may not like interpreting complex mathematical jargon, it is more beneficial to explain Shapley's algorithm in plain English.
In this instance, the Shapley algorithm essentially feeds every combination of variables into the pre-trained probabilistic model. It will leave some features out here, and other features out there; and after it has tried every possible combination of features, it is able to measure how the presence, absence or variance of certain features affects the output of the model. Utilizing this measurement—a numerical value referred to as a Shapley Value—, we can then derive an understanding of how each feature affects the model and utilize that effect as insight for future predictions.
One way to visualize this is to think of the probability model as a factory. We feed different materials (features) into our factory, and based on those materials, the factory will output a certain product (probability to convert). Obviously, the final product will not be the same if we don't give it the same materials. Shapley's algorithm uses complex mathematics to understand exactly how the materials we feed into our factory affects the output.
Take the Blinders Off
Now that we have gone into more detail about how the Shapley values are calculated, I'd like to take a step back and refocus the conversation on the task at hand: understanding the value of Google’s DDA model. It’s important to note that all of that complex math and model building going on in the background is being calculated on data from our specific GA4 instance. So the numbers for Adswerve's GA4 property will be different from the numbers from your GA4 account because each of our users behaves differently. This is where the DDA becomes so valuable. It is an attribution model that reflects your specific business and attributes values to conversions based on the Shapley values calculated as a result of your specific data. Isn't that amazing?
Now What?
With all of this exciting attribution knowledge, you may want to know how to use any of these models. To adjust the attribution model for your GA4 property, navigate to Admin > Attribution Settings (under the Property column) and select the model you would like to use under the "Reporting Attribution Model" drop-down menu.
Furthermore, Adswerve can facilitate building even more customized attribution models for you! Perhaps you have your own first-party data that you want to be included in your attribution model or maybe there are some customized windows you'd like to use. Adswerve can help! Please contact us today and we can help you build a completely unique, adjustable and powerful Data-Driven Attribution model to use in the Google Marketing Platform.