Google Analytics App+Web: Purchase and Churn Probability

July 12, 2020 - Written by

Google has got us used to quietly launching new features and metrics. Yesterday two new predictive metrics were introduced to App+Web properties: Purchase Probability e Churn Probability. Besides, the purchase probability allows to create new audiences in Google Ads.

Let’s dive in and see what’s all about.

Purchase Probability

This metric isn’t anything new as it is inside the report “Audience” if you’ve already set up the e-commerce tracking and you meet a certain threshold of transactions. Now however the App+Web Purchase Probability predicts the likelihood that active users in the last 28 days will purchase in the next 7 days. Currently the new metric is available for e-commerce purchase event and in-app purchase (they are collected via Firebase). Machine learning process defines a probability model set on the 28 previous days to the first visit of a user.

Churn Probability

The churn rate, by definition, is the percentage of users that discontinue subscribing to a service in a given period of time compared to the global number of clients that have made use of it in the same period.

The churn is one of the pivotal metric when it come to e-commerce or lead generation, because it is inversely proportional to the retention rate. Together with LTV and CAC, they show the soundness of the marketing strategic actions we are taking.

From an analytics perspective, churn probability is defined as the probability an active user who visited your app or site in the last 7 days will be most likely to be doing it in the next 7 days. Also in this case the underlying model envisages an analysis based on the 28 previous days.


The activation of the predictive metrics is conditioned by:

  • A minimum number of 1000 positive and 1000 negative cases of users who purchased or abandoned the purchase process.
  • Time range during when the 2000 cases occur (not defined but presumably of 28-30 days – as well as for Universal Analytics)
  • If the first two rules are met, Google Analytics will initialize the predictive model once a day for each user; if either one fails, Analytics will suspend the prediction updating and we could have a “not set”.

Where to find the metrics

Firstly you have to set up an App+Web property and e-commerce and/or in-app purchases tracking.

If the requirements above are met, you can find the metrics inside the Audience Builder e Analysis report.

Predictive Audiences

The new metrics, and specifically the Predictive Probability, allow to build Predictive Audiences right inside Analytics and import them into the connected Google Ads account.

As a default Google Analytics App+Web suggests the 4 audiences you can find in the screenshot below:

Regarding remarketing and retargeting the audiences can be useful to build more and more targeted ads to engage users and encourage them to convert (purchase) or to re-engage users that are likely to be less interested in our products/services (churn).

Differently from now, in the first case we are going to re-engage users that maybe never selected an item but who, via the predictive model, could purchase in the future. The audience thus is automatically able to identify which users’ actions in the website or app lead to a purchase!

And what about Google Analytics?

In addition to building audiences you can exploit the predictive metrics inside the “Analysis” report. A “killer combo” could come from prediction metrics in association with the super new life time value metric (I will talk about it in a specific post). It could be interesting to set up a test to decide how to reallocate the campaign budget considering the purchases/look-alike conversion probability.

Introducing the new metrics Google is even more pushing for “App+Web” as the new standard measurement model (although currently is not so much in terms of ease of use/comprehension). And what about you? How would you take advantage of the new metrics? Do you think they are useful? Let me now in the comments.


Leave a Reply

Your email address will not be published. Required fields are marked *

  • Analytics Boosters

    Vicolo Buranelli, 2
    31100 Treviso

  • Analytics Boosters
    United Kingdom

    41 Devonshire Street
    W1G 7AJ
    United Kingdom

  • Analytics Boosters
    United States

    80 S.W. 8th Street
    33130 - Miami
    United States of America


Contact us