The retention feature is built for customer event data. If you have your user behavior data in your database such as purchase, you can create use retention features in order to see how often your users are purchasing products on your website. We basically build a cohort table that shows you when your users churn and get back to your website if you have customer event data in your database.

Mode

The retention modes are being used to construct the funnel queries and the user selects one of them depending on the semantics of the funnel query. The support accurate and approximate modes in the funnel feature. Since the retention SQL queries are complex, you can start running the queries in approximate mode and switch to accurate if you need the exact values. approximate mode makes use of approximation and provides > 90% accuracy.

First and Returning Step

In the example above, all the users who did Appopened will be in our cohort and each of them, we check if the user did Serverconnected event for the next 14 days so that we can find out when they churn or stick in the app. You can add filters for the event types if you want to analyze only a subset of users.

Dimension

If you don't select the dimension on the retention page, you will see the data grouped by the window value (daily in the example above). In some cases, you may want to compare different user segments by drilling into the retention by comparing different values across one dimension such as the gender or location of the user. For example; if your male users are converting much less than the females, you usually need to personalize your pages for your male users in order to be able to convert them.

Holding Constant

By default, we analyze all the events done by the individual users in a funnel but in some cases, you may want to calculate the retention within the device id or session id so you can use different dimensions as holding constant.

Window

You can select different date groups if you want to analyze the cohorts for long period of time and bucket the values into days, weeks or months.

Actions

SQL: If you're technical, you can click Export to SQL under the menu Export and see the underlying SQL query of the report.
Export data: You may wish to export the data as CSV or download the image of the chart. We limit the rows when running the query and allow you to export all the data in case the count of distinct dimension values is more than 5000.

By default, we cache the data when you run queries in the retention page so if you run the exact same query twice subsequently, we show you the cached data. You can click the light bulb and force refresh if you wish.