Home Blog Google Analytics APP ...

Google Analytics APP + Web: Purchase and Churn Probability

Filippo Trocca

written by Filippo Trocca, Principal at 3rdPlace

As promised, presenting the new developments for Google Analytics or Google Analytics App + Web, Google bets on machine learning in all its products!

On July 10th, Google announced on its official blog the integration of two new models in the Google Marketing Platform: Purchase and Churn Probability.

Purchase Probability is designed for eCommerce. It helps you identify users with the highest probability of conversion within the next 7 days on your App or website.
On the other hand, Churn Probability, as far as I am concerned, is designed for publishers. It helps you identify active users, i.e. users who in the past have shown to appreciate your digital properties but they have low probability to come back to your site or app in the next 7 days.

Google Analytics models: Purchase and Churn Probability

Documentation is very sparse so far. Anyway, it seems that the models have two different declinations, as we can see in the screenshot below.

The first one, Purchase Probability, has the aim of picking up those categories of users:

  • users who could buy in the next 7 days
  • users who could make their first purchase in the next 7 days

The second, Churn Probability, is addressed to those categories:

  • users who have already made a purchase on your digital properties, but they are likely not to return in the next 7 days
  • users who will not return to your site in the next 7 days

Obviously, these models require a lot of traffic to be activated and to be effective. As we can understand reading the documentation, they are truly usable only in sites that have a lot of traffic. This is true because models are based (they learn, indeed) on user activities in the last 28 days. For model training, it is necessary to have 1000 users who completed a purchase (or who returned to the site often for Churn Probability) and 1000 users who have not met these requirements.

Consequently, in order to train the model, a digital property needs at least 2000 targeted users in the last 28 days. Among these, a subset it selected: a remarketing list to be activated must contain at least 1000 users, but any Digital Advertiser knows that this issue is far from reality of a successful remarketing campaign.

Making it short, the mail limitation of these models is the traffic needed. The second is the data itself: the model can only be used on data that Google knows and classifies. What does it mean? It can use only basic data of App and/or Web, it cannot use events, custom dimension, or metrics. It does not know how to manage them because it does not know their meanings.

3rdPlace’s DataLysm solution

As we have seen so far, Google models are very powerful, but they have some important limitations. For all these reasons, 3rdPlace has developed DataLysm.

DataLysm is a solution that, starting from the same data model as Google Analytics 360 or Google Analytics App + Web, or using its own data model, develops probability models (propensity to purchase and churn) using an huge number of data sources.

Just to cite some example, DataLysm allows:

  • customization of the tracking strategy such as custom dimension, custom metric and events
  • integration with corporate CRM
  • integration with offline activities
  • retrospective userID

This allows to enhance company’s proprietary data (known as first party data) and to extract big value from then. Moreover, the solution allows to develop the Customer Lifetime Value using all the transaction information available within the company and not only those of Google Analytics.

In particular, the restrictive userID allows not only to anonymously recognize the sessions in which the user has logged in, but it allows to analyze his behavior even when he has not explicitly logged in on your platform.

DataLysm also solves the limitation related to the number of remarketing segments, making available a conversion probability index (this index can be useful for all types of companies, not only for eCommerce) at campaign-level, ads-level and keyword level. This index allows to optimize campaigns not only in relation with the ability to generate conversions, but also analyzing the quality of generated traffic.

The integration of all these data and technologies allows Datrix Group to stand out from the market, as shows the recent business case (Euronics, supported by Bytek), picked up by Google itself.