Catch up on everthing we announced at this year's Firebase Summit. Learn more

Firebase Predictions

Firebase Predictions applies machine learning to your analytics data to create dynamic user segments based on the predicted behavior of users in your app. These predictions are automatically available for use with Firebase Remote Config, the Notifications composer, Firebase In-App Messaging, and A/B Testing. You can also link your app's Predictions data to BigQuery so you can get daily exports that you can further analyze or push to third party tools.

When you use Predictions with Remote Config, you can increase conversions by providing a custom experience based on each of your users' anticipated needs.

You can also use Predictions with the Notifications composer to deliver a one-time or recurring campaign. For example, to automatically send a notification to users who become predicted to stop using your app.

With A/B Testing, you can compare the effectiveness between different the Notifications composer campaigns, or use Remote Config to test the result of different ways to customize the in-app experience for users in a predicted segment.

Get Started

Key capabilities

Bring the power of Google's machine learning to your data Firebase Predictions applies Google's expertise in machine learning to your analytics data, creating dynamic user segments based on the predicted behavior of users in your app. With Google's powerful machine learning, you can make product decisions based on predicted behavior, rather than historic behavior.
Boost conversions through customized user experiences Firebase Predictions is integrated with Remote Config and Firebase In-App Messaging, letting you customize a user's experience based on their predicted behavior. For example, for users who are predicted to spend, you can show a new in-app purchase bundle, while for users who are predicted to not spend, you can adjust the frequency of ads. Predictions update dynamically every day, so you can always offer a fresh, personalized experience to your users.
Increase retention with smarter notifications Re-engaging a user who has already stopped using your app is tough. By using Predictions, you can engage users who are predicted to churn, before they ever stop using your app. You can set up a one-time campaign or automate sending notifications for certain predicted groups through recurring campaigns.
Create custom predictions In addition to the built-in predictions—will churn, will not churn, will spend, and will not spend—Firebase Predictions allows you to create predictions based on any conversion event you have defined in your analytics data. Once you define the event, Predictions creates a dynamic user segment composed of users who are predicted to complete that event in your app in the near future.
Export to BigQuery You can schedule automatic daily pushes of your app's prediciton data to BigQuery for further analysis or to push it into third party tools and services.

How does it work?

Predictions are available for iOS, Android, Unity, and C++ apps that include the Analytics SDK. Predictions creates dynamic segments of users who are likely to complete a certain event. You can use these segments to target users with Remote Config, Firebase In-App Messaging, and the Notifications composer.

By default, Predictions provides two types of predictions: churn, which helps you identify users likely to stop using your app (that is, they will not open the app or app-related notification messages), and spend, which helps you find users who are likely to spend money in your app. You can also create your own predictions based on custom conversion Analytics events that you collect in your app.

Predictions lets you adjust the risk tolerance of a prediction so that you can strike the right balance between segment size and accuracy. With Low risk tolerance, you target fewer users with more accuracy, or you can use High risk tolerance setting to get more of the users you want but at the risk of including more false positives.

The machine learning model for your app improves as the amount and relevance of data collected using Analytics increases, and as your number of users increases. In addition, the accuracy of the model for a specific user will improve further after that user has used the app for at least a few days.

Do I need to share my data with Google in order to use Firebase Predictions?

You are not required to share your Google Analytics data with Google to help improve Google's products and services. You can turn that off in the Firebase console by going to Analytics-> Dashboard-> Settings at any time you wish.

You do need to use Google Analytics to log your app data to Firebase. You also need to make sure your Google Analytics data is available in Firebase (see Data-sharing settings).

Please note that Google Analytics uses a shared model. While your raw event data is safe and only available to you inside Firebase, the model quality does improve for everyone who uses Predictions. You need to explicitly opt into Predictions from the Predictions homepage. You can turn off Predictions at any time and your data will no longer be available for Predictions even if it is still available in the rest of Firebase.

Implementation path

Predictions works best for apps with 5K or more monthly active users. It also works best for apps with frequent usage and spend such as games.

Add Analytics to your app To make predictions, your app needs to record events using Google Analytics.
Enable Predictions and monitor prediction readiness Use the Firebase console to start making predictions based on your app's analytics data and to monitor whether predictions has enough data for the built-in churn and spend predictions. You can also use the Firebase console to monitor whether enough Analytics data is available for predictions that you create based on additional Analytics events collected by your app.
Access predictions results programmatically from your app

From the Firebase console, create a Remote Config parameter that returns the result of a given prediction for a given users. Once you have the Remote Config SDK integrated into your app, you can call that Remote Config param programmatically from your app to know the result of a specific prediction for a specific user.

For example, you can offer a different first screen experience for the user who are more likely to spend.

Reach users in a certain prediction segment using the Notifications composer or Firebase In-App Messaging

You can contact users in your predicted user segments using one-time or recurring campaigns using the Notifications composer or Firebase In-App Messaging.

For example, you could use the churn prediction with high risk tolerance to automatically send notification messages to new users who become likely to churn or stop using your app.

Next steps