Churn prediction

The churn measures an app loss in payments, launches, and other audience activities for a given period. Users can stop using an app as early as the next day or a month after installation.

Predictions allow you to analyze audience engagement and prevent user drop-off. With forecast, you can:

  • get an insight into a churn likelihood in specific audience segments, for example in a particular country, and reengaged those users;
  • identify campaigns with dismal forecast and rebalance ad budget;
  • find out ads that are better attracted to a loyal audience and focus resources on them.

How it works

MyTracker makes a forecast using machine learning models. Models analyze captured app data: installs, launches, payments, sessions, and can make a forecast one day after a user installed your app.

Churn rate prediction

Churn can be forecast for periods of 7, 14, 21, or 28 days after app installation. Predictive churn does not mean the app uninstall, it is a forecast about a user will not be active during a specified inactivity window ? following a selected forecast day.

For instance, 20 users installed your app on 2021.01.01. Churn Prediction LT 7d = 50% would mean 10 users will not be active for the period from 2021.01.08 till 2021.02.07 (during 30 days of inactivity window).

Churn probability prediction

In addition, MyTracker predicts the churn probability for the 7th day after installation for each user.

For instance, 20 users installed your app on 2021.01.01. The forecast will show you which of these 20 users will not be active after 08.01.2021 (during 30 days of inactivity window) with a probability greater than 80%.

How to use

If you have the churn prediction, you can take actions to retain users: relaunch ad campaigns, send push notifications, use deep links or personalize offers.

Churn rate prediction

The churn prediction shows a percentage of devices that are probably going to stop using your app. Use this prediction to analyze potential losses for individual ad campaigns and countries:

  1. Make sure that the MyTracker SDK has been integrated into your app.
  2. Check inactivity window settings on your Project (30 days by default). This is a period of user inactivity used for the forecast.
  3. In the Builder, select a report period. The forecast will be made for the cohort that installed the app during the selected period.
  4. Add Churn Prediction metrics to your report. Churn can be forecast for periods of 7, 14, 21, or 28 days after app installation.
  5. Add dimensions. For example, Select from list → Traffic source → Campaign (the forecast demonstrates the best quality in the Campaign section) and Select from list → Geo&Demography → Country. Press the Calculate button.

Churn prediction LT metrics show the percentage of devices on which users installed your app during the reporting period and is forecast to be inactive after the selected period (7d, 14d, 21d, 28d) based on the inactivity window settings.

You can get reports with Churn forecast using Export API

Churn probability prediction

Churn probability prediction allows you to get a list of users, who will stop using the app with a probability of 80 – 100%. By using this list, you can run targeted campaign and reduce churn.

To download device identifiers with a high churn probability:

  1. In MyTracker, create the segment of devices:
    • Select Devices in the Audience type field.
    • Edit the Period parameter. The exported file will include devices with installations made in the selected period.
    • Add the Churn Prediction parameter and specify the churn probability for which you need to export the data. For example, add an interval of 80 – 90 to export devices, that have an 80 – 90% churn probability after 7 days after installation.
    • Add the Partner or Campaign parameter for which you need to export the data.
  2. For details, refer to the Add segment section

  3. Get access to the Export API (see the Access to API section).
  4. You can export the data using the MyTracker interface if you only need to get the IDs of the devices without additional data about installation time, campaign, etc. For details, refer to the Export segment section.

  5. Create a request to export raw data on event installs:
    • Specify idSegment created in the first step. You can find it on the segment page in the address bar of your browser.
    • Add selectors that need to export. For example, a device identifier idfa, advertisingId, time of installation tsEvent. See the complete list of selectors in the Selectors dictionary.
    Example query:
  6. /api/raw/v1/export/create.json?event=installs&selectors=idApp,idfa,advertisingId,tsEvent&dateTo=2022-02-28&dateFrom=2022-02-01&idSegment=1111&timezone=Europe/Kiev

    For details, refer to the Export segment section

You will get a CSV file with list of device identifiers with the specified churn probability.