Retention Analysis

Video tutorial

Retention analysis video tutorial

Retention Analysis Overview

Retention analysis is an analytical model used to analyze user engagement/activity levels, examining how many users who perform an initial action will take subsequent actions. This is an important metric for measuring the value of a product to its users.

Retention analysis can help answer the following questions:

• Did a new customer complete the actions you expected users to perform within a certain period of time? Such as paying for an order.

• A social product has improved the onboarding process for new registered users, hoping to increase user engagement after registration. How can this be verified?

• To determine whether a product change is effective, such as adding a feature to invite friends, observe whether anyone uses the product for several more months as a result of the new feature.

View examples of retention analysis features.

Retention Analysis Interface Function Introduction

Select initial and subsequent behaviors

There are two strategies for choosing the initial and subsequent actions:

1. Initial actions are selected based on events that users trigger only once, such as "register," "upload avatar," or "activate device." Subsequent actions are selected based on events that you expect users to trigger repeatedly, such as "read articles," "post," or "purchase." This retention strategy is used to compare and analyze the engagement of new users at different stages of product adoption, thereby evaluating the gains and losses of product iterations or operational strategy adjustments.
2. The initial and subsequent behaviors are chosen to be the same, and the event is expected to be triggered repeatedly by the user. This retention method is used to analyze the usage patterns of loyal users.

Set initial and subsequent behavior filtering conditions

Based on the attributes of an event, you can filter the initial behavior or subsequent behavior into specific dimensions according to your needs. For example, if we want to analyze the retention rate of users in Beijing who register and then purchase mobile phones, we can define the initial behavior as "register" and add the filter condition "city equals Beijing", and the subsequent behavior as "product details of payment orders" and add the filter condition "product type equals mobile phone", which will satisfy our analysis needs.

Add metrics to be displayed simultaneously

When performing retention analysis, you can add metrics to display simultaneously, and calculate the metric values for retained users. For example, you can calculate the total recharge amount, average recharge amount, cumulative total, cumulative average recharge amount, cumulative cumulative total, and cumulative average recharge amount for each stage. Currently, simultaneous display of attributes in grouped categories is not supported.

The LTV (Lifetime Value) can be calculated using the cumulative average value per user over a specific period, based on the cumulative average value per user who visited the site on a specific date . The system displays the user's LTV in a table by selecting the revenue event as the event and the monetary attribute as the metric.

View by attribute

The retention table is grouped by the initial action date by default. The first column of each row represents the initial action date; the second column is the total number of users (unique users) who triggered the initial action on that date; the following columns show the number of users who triggered subsequent actions after the corresponding time, and the percentage of users who triggered the initial action. If other metrics are also displayed, their corresponding values will be shown.

In addition to grouping by initial behavior date, you can also group by initial behavior event attributes or subsequent behavior event attributes. For example, if you choose to group by registration channel based on initial behavior event attributes, you can see the subsequent retention data for different registration channels.

It's important to note that when setting associated attributes is enabled, grouping and viewing are only supported based on the associated attributes of the initial event. For example, if you select the page title as the associated attribute for both the initial and subsequent behavior events, you can see the subsequent retention data for different pages.

Find it a bit complicated? No problem. Hover your mouse over each cell, and a text tooltip will tell you the specific meaning of that cell. Additionally, the background color of the cells visually reflects the retention rate.

• If the selected attribute here is a numeric type, you can customize the grouping interval. If not set, the query engine will dynamically calculate the grouping interval. This setting only applies to the current query and will also apply to bookmarks after the query is saved.

• If the selected attribute is a list, all attribute values in the list will be split into independent sub-items, allowing you to view the retention analysis under each individual attribute. Note: If multiple list groups are set, only the last list type will be displayed.

Set user filter criteria

Filter appropriate analysis targets based on user attributes, user segments, and user tags. For example, only examine the retention rate of female users.

• User attributes with a time type support setting time ranges relative to the initial behavior event time and the subsequent behavior event time. Comparison methods include: before, after, today, today, and today. Time granularity supports days, hours, minutes, and seconds. For example, setting a user attribute: first login time, relative to the initial behavior event: App login within the same day, will select users who logged in on their first day.

• User segmentation and tagging: Supports configuring user segmentation and tagging results. You can select a fixed segment or dynamically match based on the occurrence time of the initial behavioral event.
  
  • Latest results: The selected users are the latest results for each group.
  • Custom: The filtered users are the calculation results for a specific selected subgroup.
  • Matching by initial behavior time: Matching can be performed based on the date of the initial behavior and the date of user segmentation. For example, if the initial behavior event occurred at 11:01, it can dynamically match the segmentation results for 11:01, allowing for user selection. This is suitable for retention analysis of dynamically segmented user groups.

Set associated attributes

It supports associating attributes between initial and subsequent behavior events. Attributes associated with different events can be the same or different, but the attribute types must be consistent. For example, a content website might want to know the seven-day retention rate of each page, therefore it needs to set attributes for the initial behavior event. and subsequent behavioral events App 浏览页面 Adding a related attribute, such as the page title, will then associate the page with that attribute, ensuring that users are matched strictly according to this pattern.

Select the time period and retention/churn duration for the assessment.

The time range selected here is the time range in which the initial behavioral event occurred. As shown in the figure above, if "7-day retention" is selected, the cutoff date of the subsequent event time range will be extended to July 12, 2020 (extended by 7 days from July 5, 2020).

You can view retention/churn data for different time periods by day, week, and month. Version 1.6.5 provides a function to view churned users; click the area circled in red in the image above to select the churn time.

In retention analysis, a churned user is defined as someone who has not engaged in any subsequent events for several consecutive days (excluding the day itself, i.e., starting from day 1 after the initial behavioral event is triggered). For example, in the above figure, "day 2" churn means that 2,377 people did not engage in any subsequent events for two consecutive days after engaging in the initial event (regardless of whether the user triggered any subsequent events on June 30th).

Retention Form

In the retention table, hovering the mouse over the table will display relevant tooltips. Clicking on the number of users allows you to view the user list and create user groups.

How do you calculate the overall retention rate over a period of time?

As shown in the figure below: The overall retention rate on day 4 in China is the "weighted average" of the retention rates on day 4 of the two initial behavior dates, 2021-04-05 and 2021-04-06.

Note: The weighted average is calculated as the percentage of the number of people who initially acted on the current day compared to the total number of people who initially acted. For example, the weighted average for April 5, 2021 = number of people who initially acted / (number of people who initially acted on April 5, 2021 + number of people who initially acted on April 6, 2021) = 144 / (144 + 163)

Therefore, the overall retention rate on the fourth day = 34.03% * 144 / (144 + 163) + 20.25% * 163 / (144 + 163) = 49 / (144 + 163) + 33 / (144 + 163) = (49 + 33) / (144 + 163)

In SA versions 2.1 and later, the number of users in the retention/churn rate of the summary row (a table row that supports collapsing and expanding) is not deduplicated; it is simply the sum of the retention/churn users for each time period below divided by the sum of the total number of users for that time period . When viewing by "Overall," you can choose to "Show Other Metrics Simultaneously." The metrics shown simultaneously will also be displayed in the table.

View user details

The numbers in the table cells are clickable; clicking them allows you to view detailed information about these users and further explore the detailed behavioral sequences of individual users.

Retention Trend Curve

Use another visualization method to show the comparison of retention rates for different groups.

How is retention calculated?

The numbers displayed in retention analysis represent the number of unique users. This indicates how many users who performed the initial action within the selected time frame performed a subsequent action on the nth day/week/month.

Basic calculation rules

Suppose the initial behavior is event A, the subsequent behavior is event B, and the filtering period is from January 1, 2015 to January 8, 2015. Note that this time range is the time range during which event A occurs, and the time range during which event B occurs is from January 1, 2015 to January 15, 2015 (January 8 plus 7 days).

No associated attribute set

The table below shows the actual behavior sequence and the behavior sequence included in the calculation for a user from January 1, 2015 to January 8, 2015 (retaining the user's first initial behavior event or subsequent behavior events of the day). In the table, letters A and B represent events, and numbers 1, 2, and 3 represent the attribute values of a certain attribute of the user.

The user's retention rate is shown in the table below:

1. Without grouping, if the initial behavior date is specified as January 1, 2015, then the user is a retained user on day 1, day 5, and day 7 respectively. If no specific initial behavior date is specified, then the user performed initial behavior on January 1, January 3, January 4, January 5, and January 7 respectively, and the subsequent retention days are day 1 (initial behavior date: January 1, January 5, January 7), day 2 (initial behavior date: January 4), day 3 (initial behavior date: January 3, January 5), day 4 (initial behavior date: January 4), day 5 (initial behavior date: January 1, January 3), and day 7 (initial behavior date: January 1).

2. Grouping by the attributes of the initial behavioral event A: If users complete event A with different attribute values, the user will only be grouped into the group with attribute value 1 for the first event A that occurred on January 1st (Note: Although the user had an event with attribute value 3 on January 4th, the user will still be counted in the group with attribute value 1). This user had initial behavioral event A on January 1st, January 3rd, January 4th, January 5th, and January 7th, respectively, making them a retained user on days 1, 2, 3, 4, 5, and 7 under the group "Event A with attribute value 1".

3. Grouping by the attributes of subsequent behavioral events B: If users complete different attribute values for events B, the user will only be grouped into the group with attribute value 2 for the first event B that occurred on January 2nd. This user performed the initial behavioral event A on January 1st, 3rd, 4th, 5th, and 7th respectively, and is a retained user under the group "Attribute value of event B is 2" on days 1, 2, 3, 4, 5, and 7.

4. Group by user attributes, such as by gender. If a user is female, then the user is a retained user on day 1, day 2, day 3, day 4, day 5, and day 7 with the attribute value "female".

Set associated attributes

The table below shows the actual behavior sequence and the behavior sequence included in the calculation for a user from January 1, 2015 to January 8, 2015 (one initial behavior event or one subsequent behavior event with different attribute values are retained on the same day, and the first one is retained). In the table, letters A and B represent events, and numbers 1, 2, and 3 represent the attribute values of a certain attribute of the user.

Grouping by attribute of initial behavioral event A: If grouped by attribute value 1, only initial or subsequent behavioral events with attribute value 1 will be included in the calculation. These users are the retained users on days 1, 2, 3, 4, 5, and 7 for attribute value 1. If grouped by attribute value 2, only initial or subsequent behavioral events with attribute value 2 will be included in the calculation. These users are the retained users on days 1, 5, and 7 for attribute value 2. Similarly, these users are the retained users on days 3, 4, and 5 for attribute value 3.

Meaning of filtering conditions

Like other analytics features, retention analytics also offers filtering capabilities. Retention analytics provides two different filter types.

1. User attribute filtering: For example, if we add a filter condition that the "gender" is "male", then only users whose "gender" attribute is "male" will meet this filter condition and appear in the retention analysis results after filtering;
2. Event attribute filtering: Similar to the triggering restrictions of the funnel, this filter specifies that an event must satisfy a specified attribute.

The meaning of grouping

Retention analysis offers two different grouping types. We will explain in detail using a 7-day retention period from January 1st to January 8th, 2015, where the initial behavior is A, subsequent behavior is B, and the time frame is:

1. Grouping by User Attributes: Further grouping based on user attributes. For example, if we add a grouping condition of "gender", then the retention analysis results will be grouped into "male" and "female" respectively.
2. Grouping of event attributes: For example, if we choose the initial behavior attribute "screen height" as the grouping setting, this grouping means that within the time range of January 1st to January 15th, 2015, events will be grouped according to the value of the initial behavior attribute "screen height". Below are some specific examples:
   1. A user's behavior sequence during this time period is A, B, C, A, B. The "screen height" value of the first occurrence of A is "320", and the "screen height" value of the second occurrence of A is "1080". Because the grouping is based on the "screen height" of the first occurrence of A, this user will be classified into the statistical results of the "320" group.
   2. A user's action sequence during this time period is A, A. This user has no subsequent actions after the initial action A. The "screen height" value of the first occurrence of A is "1080", and the "screen height" value of the second occurrence of A is "320". Because the grouping is based on the "screen height" of the first occurrence of A, this user will be classified into the statistical results of the "1080" group.

FAQ

Why do we need to do retention analysis? Isn't looking at the percentage of active users enough?

Retention analysis grouped by initial behavior time can eliminate the impact of user growth on user engagement data. If a product is currently in a rapid growth phase, the increase in active users among new users may mask changes in the activity of existing users. Through retention analysis, you can segment users by registration time and draw conclusions such as: "Before the March update, the 7-day retention rate for users who registered that month was only 15%; however, after the April update, the 7-day retention rate for users who registered that month increased to 20%." Similarly, retention analysis based on non-time dimensions has similar value; for example, it can show whether the launch of new features has different effects on the retention of users of different genders. When analyzing user retention, we must identify valuable follow-up behaviors based on actual business needs and analyze the value retention of users to provide substantial guidance for product optimization and improvement.