About predictive metrics
Google Analytics automatically enriches your data by bringing Google machine-learning expertise to bear on your dataset to predict the future behavior of your users. With predictive metrics, you learn more about your customers just by collecting structured event data.
| Metric | Definition |
|---|---|
| Purchase probability | The probability that a user who was active in the last 28 days will log a specific key event within the next 7 days. |
| Churn probability | The probability that a user who was active on your app or site within the last 7 days will not be active within the next 7 days. |
| Predicted revenue | The revenue expected from all purchase key events within the next 28 days from a user who was active in the last 28 days. |
Currently, only purchase/ecommerce_purchase and in_app_purchase events are supported for the Purchase probability and Revenue prediction metrics.
ecommerce_purchase event, we now recommend the purchase event instead.Prerequisites
To train predictive models successfully, Analytics requires that the following criteria are met:
- A minimum number of positive and negative examples of purchasers and churned users. In the last 28 days, over a seven-day period, at least 1,000 returning users must have triggered the relevant predictive condition (purchase or churn) and at least 1,000 returning users must not.
- Model quality must be sustained over a period of time to be eligible. (Learn more about what actions you can take to make sure your property has the best chance possible of being eligible for predictive metrics.)
- To be eligible for both the purchase probability and predicted revenue metrics, a property has to send the
purchase(recommended for collection) and/orin_app_purchase(collected automatically) events. When you collect thepurchaseevent, you need to also collect thevalueandcurrencyparameters for that event. Learn more about purchase events.
Predictive metrics for each eligible model will be generated for each active user once per day. If the model quality for your property falls below the minimum threshold, then Analytics will stop updating the corresponding predictions and they may become unavailable in Analytics.
You can check the eligibility status of each prediction by going to the predictive section within suggested-audience templates in the audience builder.
Using predictive metrics
Predictive metrics are available in the audience builder and in Explorations.
Audience builder
Predictive metrics can be used to create predictive audiences in the audience builder.
Exploration
You can use Purchase probability and Churn probability in Explorations within the User lifetime technique.
Not all users have the same data available. This means when you use a query that includes all users in Google Analytics 4, you'll see results broken down into 2 groups:
- Users with prediction metrics: This group includes users for whom Google Analytics 4 can calculate things like purchase probability.
- Users without prediction metrics: This group includes users for whom Google Analytics 4 doesn't yet have enough data to calculate predictions.
Best practices
In your data-sharing settings, enable the Modeling contributions & business insights setting. You benefit when this setting is ON because Analytics is able to use shared aggregated data to improve model quality and improve your predictions.
Make sure to maximize the use of event recommendations in your property.
Make sure you are collecting the purchase, and/or in_app_purchase events. in_app_purchase events are collected automatically. However, you must link to Google Play via your Firebase account in order to view the in_app_purchase event if you have an Android app. Keep in mind that although we will continue to process the ecommerce_purchase event, we now recommend the purchase event instead.
If you define a custom audience and add predictive conditions to use In-app purchase probability and Purchase probability, only users who complete both a purchase and an in_app_purchase will be included in the audience.
Collecting a larger variety or volume of recommended events corresponding to user behavior will help enhance our models and improve predictions. Likewise minimizing noisy events that are not meaningful in terms of user behavior will also help improve predictions.