Acquiring new users requires a significant amount of resources to drive the action of downloading and installing your mobile app. So after much time and effort (and money), maybe your app enjoys high download rates and you congratulate yourself on a job well done. But don’t sit back and relax yet. Because while thousands of users are downloading your app, how many of them actually continue to use your app over time? It's a worthy question to ask if you consider that in almost all cases, it is far cheaper to retain an existing user than to attract and register a new user.
By looking at retention, you can create the opportunity to grow a devoted user base and foster a reliable source of revenue. With multiple methods of increasing user retention (essentially, focusing on keeping and engaging your existing users), you first need to understand the behavior of the existing users of your app, and their lifecycle from download to continued use or disuse of the app.
The value of retention reporting isn't only about assessing user loyalty over time, but having the insight and quantifiable data to see when and where retention starts to decline (so you know where to target your resources more precisely, effectively, and efficiently). You can also use filters and segments to view the same data from different angles (for example, see which specific ad networks and advertising partners have the highest retention rates), then adjust your ad spend accordingly. Ultimately, retention is not about getting a high quantity of users by driving installs through mobile advertising campaigns, but being able to identify your high-quality users so you know which ones to nurture (for example, through cheap email marketing campaigns).
Definition of Retention
While user acquisition focuses on driving new installs (acquiring new users), user retention focuses on keeping existing users (who installed your app). In analyzing retention, you can see how long users continue to get value out of your app (by performing in-app events, completing purchases, or accomplishing whatever behavior you deem valuable). The longer users find value in your app, the better your retention rates are, which ultimately leads to higher user lifetime value (LTV) for you.
So retention is the ability of a product or service to retain its existing customers, with retention rates measuring the ratio of customers who are devoted and “stay” (those who keep using and supporting your app) to the total number of customers who installed your app. In contrast, user attrition is the opposite case where you measure the ratio of customers who are dissatisfied by the app and “leave” (stop using and supporting your app) against your total install base for the app.
Retention Rates
Measuring the rate of retention provides a useful metric for understanding user behavior and the complete user lifecycle. Retention rates typically measure the number of active users (unique users who completed a desired action regardless of how many times they did so) as a percentage of the entire user cohort to which they belong (for example, all users who installed your app 30 days ago) over time.
In Attribution Analytics, our cohort-based retention analysis measures the ability of a mobile app to retain its install base. Attribution Analytics does not measure attrition because it is essentially the inverse or absence of retention. Cohort-based retention analysis consists of two main components:
- Cohorting event - the initial engagement event in which you want to aggregate users together, which requires defining the type of event (typically the app "install" event), the timeframe of interest (during which the event occurred), and the cohort time interval by which users are grouped (for example, a weekly interval).
- Retention indicator metric - the event used to determine the retention rate itself, as a function of the cohorting event. This metric in Attribution Analytics is unique app opens by users.
Attribution Analytics calculates retention rate by dividing the number of unique app opens (within a specified time interval) by the total number of app installs (essentially, the base cohort size or retention potential), and then multiplying the quotient by 100.
| Retention on Day X = ( | ) X 100 | |
The denominator (total installs) defines the cohort and its size, and is bound by a time range and interval so that you can evaluate cohorts with one another for an "apples-to-apples" comparison. Defining cohorts by time range and interval (limiting the analysis to only the installs performed within a specific time range) allows Attribution Analytics to calculate retention (number of unique app opens) relative to the initial install date and time.
The numerator can vary based on your definition of a "retained user". The two main methods of calculating retention differ only by the numerator:
- Classic Retention - counts users as retained in an interval if they are active in that specific interval.
- Rolling Retention - counts users as retained in an interval if they are active in that specific interval, or a future interval.
Attribution Analytics currently uses rolling retention, but let's take a look at both methods for comparison.
Classic Retention
Classic retention generally provides a conservative calculation of retention because it only considers the portion of users retained on a given day (it does not account for future activity). Classic retention is best for weekly or monthly reporting intervals because longer intervals generally allow more users to be retained within the interval.
For example, let’s suppose User 1 performed the following actions:
- On March 1 (Day 0), installed a mobile app
- On March 2 (Day 1, first day after install), was not active/did not open the app
- On March 3 (Day 2), was active/opened the app
Since the user wasn’t active on Day 1, classic retention considers the user NOT retained for that interval. Then the user was active on Day 2, so they are retained for that interval.
Now let’s assume User 2 performed the following actions:
- On March 1 (Day 0), installed the app
- On March 2 (Day 1), was active/opened the app
- On March 3 (Day 2), was not active/did not open the app
The following diagram shows the activities of User 1 and User 2 over one week, and their impact on retention levels using the classic retention method.
So with classic retention, we calculate retention as follows:
- Retention for Day 1 is 50% because only one of the two installed users was active between hour 24 and hour 47 (from Day 0)
- Retention for Day 2 is also 50% because one of the two installed users was active between hour 48 and hour 71 (from Day 0)
- Retention for Day 3 is 0% because neither users were active during that interval.
Rolling Retention
Now let’s take the same scenario as above with two users performing the same actions over one week, and look at how rolling retention is calculated (which is more complicated because retention levels may change over time as this method takes into account 30 days after the current interval). Rolling retention is best for daily reporting intervals because future activity is already accounted for in prior intervals (expanding the interval does not affect the retention rate).
Taking the same two-user scenario above, the following diagram shows their impact on retention using the rolling retention method.
With rolling retention, we calculate retention as follows:
- Retention for Day 1 is 100% because both users were either active on that day or some point in the future
- Retention for Day 2 is also 100% for the same reason
- Retention for Day 3 is also 100% for the same reason (even though neither users were active on Day 3, they were both active on a future day within the interval)
The following table compares the retention rates between classic and rolling retention.
| Classic Retention | Rolling Retention | |
| Day 1 Retention Rate | 50% | 100% |
| Day 2 Retention Rate | 50% | 100% |
| Day 3 Retention Rate | 0% | 100% |
| Day 4 Retention Rate | 50% | 100% |
| Day 5 Retention Rate | 50% | 100% |
| Day 6 Retention Rate | 50% | 50% |
| Day 7 Retention Rate | 0% | 50% |
| Day 8 Retention Rate | 50% | 50% |
As you can see, rolling retention provides a different indication of retention (generally a higher one) than classic retention even though the same two users performed the same actions over the same interval. Even if a user is not active in an interval, rolling retention may consider the user retained for that interval if they are active in a future interval. But classic retention does not account for activity in future intervals.
Only when a user is no longer active in a future interval (stops using the app) are they considered not retained. User 2 is no longer active after day 5 and thus, no longer considered retained with rolling retention. Retention for Day 6 and beyond is only 50% because only one of the users is retained (User 2 stopped using the app after Day 5).
Because rolling retention accounts for future activity (up to 30 days after the current interval), calculation of Day 1 retention is subject to change and cannot be deemed complete for 30 days (within the 30 days, a user may become active and thus affect the retention rate for a prior interval).
With classic retention, the calculation is easier because future activity does not impact the retention rate for a current interval. So taking the same scenario above, a user is considered retained for Day 1 (24-47 hours after install) if they opened the app on that day. And calculation of Day 1 retention is complete after 48 hours from install (any activity performed thereafter does not affect the Day 1 retention rate).
Please note that the Retention report provided in the Attribution Analytics platform automatically excludes testing and debugging data (installs, opens, and events associated with debugging requests or test profiles). This exclusion prevents test data from skewing the retention calculation.
Implications of Retention
Now let’s consider a big-picture example to demonstrate what retention means. You recently released a significant update to your app and you’re interested in evaluating if the update had any (hopefully positive) effect on the retention rate for your app. As the update went live two weeks ago, you can compare the users who downloaded and opened the old version of your app (during the two weeks prior to release/update) to the users who downloaded and opened the new version of your app (in the two weeks after your app went live).
Now let's say your cohorting event is defined as “First App Open/Install” beginning the week of April 1, 2014 and ending the week of April 29, 2014 (using a weekly interval). This definition aggregates all of the users who installed your app (or opened it for the first time) during the month of April into 4 weekly cohorts. The retention indicator metric is defined as “App Opens” and is measured both as an absolute count of unique users (who completed the event) and as a percentage of unique users (who completed the event, out of the entire cohort),
Using the retention report, you can see that your example app update had a positive effect on retention, as your retention rates are higher in the two cohorts post-update event (in green color in the table above) than the two cohorts of users who opened your app BEFORE you released the update (in blue color above). So the data substantiates that your app update had a positive effect on your retention rates, congrats!
While this example scenario compared the retention rates before-and-after an app update that you made (correlating user behavior and response with a change you implement), you can run retention reports to look at data in comparison to other factors or variables. For example, you might compare the retention rates between two different partners to see which one you should partner with (which one generates more valuable/high-quality users).
Example Retention Reports
The following scenarios provide examples of how you can use retention reporting in Attribution Analytics to assess user loyalty and identify potential pain points (when do users drop-off?).
Which of the six ad campaigns that I ran in April was most effective?
- Define/limit the cohort to only the installs in April
- Group the installs by ad campaign
- Filter the list of installs to only the six ad campaigns of interest
- Show 30-day retention
How did my Android users compare to my iOS users last quarter?
- Define/limit the cohort to only installs
- Group the installs by device OS
- Show 90-day retention
How well did I retain users on a per-country basis last quarter? (Should I vary my target CPI on a per country basis?)
- Define/limit the cohort to only installs
- Group the installs by country
- Filter out "organic installs" (non-attributed installs) from the list of installs
- Show 90-day retention
No comments:
Post a Comment