2023-01-03 | Article | Insights
What is Predictive Marketing? Based on predictive marketing, existing data can be applied to future user behavior (forecasts). An example is to collect tracking data on the website and then use it for various use cases for Google Ads and Display & Video 360 (DV360). Data is created based on user behavior which could be a purchase or an impending cancellation.
What general steps are involved in Predictive Audiences?
What are aspects to keep in mind when working with Predictive Audiences?
Currently, predictions can be made for purchase probability, cancellation probability, and expected revenue. More may follow in the future. Google has yet to make any further announcements. A custom model must be considered for other business metrics at the moment. In addition, buyers and churned users must be sampled in a minimum of 1000 and Predictive Audiences must meet this requirement. A general traffic of at least four weeks is also required for the model to be of good quality.
The above figure shows the Predictive Audience Builder in GA4. The following is an excerpt of the possible Predictive Audiences that can be created:
The previous table shows how Predictive Audiences are described and configured. You can also define the duration of the users' stay in the audience in the settings which will be determined by your first data strategy. You can, for example, set a list with a duration of seven days, fourteen days, 21 days or pick the maximum duration available.
In addition, it is possible to apply further settings with conditions including sequences. A specialist in Google Analytics 4 would be required for this job which Digitl can provide.
Due to the direct integration of GA4 with BigQuery that is available to both clients using the free GA4 version as well as GA4 360 you can go beyond the Predictive Audiences provided in GA4 by default. With the BigQuery integration, raw event data from Google Analytics 4 can be exported to BigQuery to create your own predictions with Google Cloud tailored to your business specifics.
For example, this is how custom Conversion Probability Predictions (CPP) can be created. The advantages of custom predictions is that a better data understanding can be gained and a higher control over how predictions are created can be applied. On the other hand, default predictions such as the aforementioned are often a black box. When custom predictions are created, the aim is to delve deeper into the data and to define and analyze the data according to a data model and then apply it specifically to advanced use cases.
In order to conduct such a project, the Cross Industry Standard Process for Data Mining (CRISP-DM), that is often used as the fundament for data science projects, can be applied. The process includes the following six main phases:
The picture illustrates how a market-known data mining model might look and be implemented. As shown, modeling is not the only aspect to consider. The process starts with building an understanding of the client’s business to then continue with a stage of gaining data understanding. It might also include recommendations for additional tracking and/ or adjustments of the existing one if whenever important data is not tracked accurately or at least not to the extent and in the way needed to fully understand relevant user behavior which is discovered by a tracking health check.
A project for building predictions and using them the create Predictive Audiences is a cross-functional project including several units: Digital Analytics for the aforementioned tasks, Data Science to work with the data and create the predictive outcome, Cloud Engineers to automate the pipelines, and the AdTech team to set up the audiences and provide advice on the activation strategy. It takes a lot more time and effort to handle the more advanced predictive use cases as they require the use of multiple solutions by Google and the aligned work of multiple units.
Predictive Audiences compared to the more traditional retargeting methods, are an excellent way to focus your campaign spend on valuable users and expand your audiences towards users with similar attributes as these. You should keep in mind that you need enough data for a good model performance. Google offers a simplified way to create such audiences and then use them in the various tools such as Google Ads and DV360 while being able to see them in SA360. If you are interested in exploring more advanced and sophisticated options, you should check out the Google Cloud solutions and think of a Data Science project. Thanks to the native integration between BigQuery and Google Analytics 4 and solutions such as the Vertex AI, more options than ever are provided. Digitl can help find a suitable solution for your business needs - either with the default options in GA4 or with a Data Science project to dive deeper into the topic.