This article will help you:
Amplitude’s Retention Analysis chart helps you drive product adoption by showing you how often users return to your product after firing an initial event. This article will describe how the lower module of the Retention Analysis chart works, and how you should interpret the data it contains.
Analyzing your retention analysis data takes place in the screen’s lower panel. There you will be able to:
- Choose between the Retention View and the Usage Interval View
- Specify the method Amplitude uses to measure retention (N-day, unbounded, or custom)
- View data for either retention or change over time
- Define how Amplitude measures days when it calculates retention (24-hour windows or calendar dates)
- Set the time frame for your analysis using the date picker
Before you begin
How time works in a retention analysis
In a Retention Analysis chart, there are two ways to think about what a “day” is: as a series of rolling 24-hour windows, or as strict calendar dates. The method you choose can affect your results, so please read this section closely.
By default, Amplitude treats a day as a rolling 24-hour window, one that is different for each user. When a user fires the starting event, Amplitude considers that to be the beginning of that user’s Day 0. In other words, Day 0 runs for the first 24 hours after the user fires the starting event. Day 1 runs from hour 24 to hour 48, Day 2 from hour 48 to hour 72, and so on.
Under this approach, each day will be exactly the same length, no matter when the user fired the starting event.
Daily retention using 24-hour windows
- Daily retention is computed on an hourly basis.
- When Amplitude calculates daily retention, it rounds event timestamps down to the most recent hour. This means an event fired at 5:59 pm will have a timestamp of 5:00 pm.
- A user is counted as "next day" retained if they perform any event during or after the 24th-incremented hour, but before the 48th-incremented hour. For instance, if a user performs their first event on December 1st at 5:59 PM, and then a second event on December 2nd at 5:00 PM, Amplitude counts that user as Day 1 retained (instead of Day 0), because that is the 24th-incremented hour after the timestamp on the original event.
- When a user fires the initial event multiple times, Amplitude will start multiple 24-hour buckets for them. This means it’s possible for one return event to define a user as both Day 1 and Day 2 retained.
Weekly and monthly retention using 24-hour windows
- Weekly and monthly retention is computed on a daily basis.
- A week is defined as seven days. If a user fires their first event on December 1st, Amplitude considers them as Week 0 retained if they fire an event on any day between December 1st and December 7th. Likewise, Amplitude will consider them as Week 1 retained if they fire an event on any day between December 8th and December 14th.
- A month is defined as 30 days. If a user fires their first event on December 3rd, Amplitude considers them as Month 0 retained if they fire the return event on any day between December 3rd and January 2nd. Likewise, Amplitude will consider them as Month 1 retained if they fire the return event on any day between January 3rd and February 2nd.
NOTE: These computation methods only apply to data collected on August 18, 2015 or later. Any retention computations including dates before that will be computed by calendar days/weeks/months.
However, you can also set Amplitude to measure retention by strict calendar dates, where day N is measured from the calendar date the event was fired.
Daily retention using strict calendar dates
- Daily calendar dates start when the calendar day starts, and end when the calendar day ends.
- The calendar view is determined by the timezone specified in your project settings. Under the strict calendar view, daily retention is based on the calendar day instead of on an hourly basis.
- A user is counted as “next day” retained if they fire any event during the next calendar day. For instance, if a user fires their first event on December 1st at 5:59 PM, Amplitude considers them as Day 1 retained if they fire an event any time before December 2nd at 11:59 PM. If a user fired their first event on December 1st at 11:50 PM, they have until December 1st at 11:59 PM to fire the return event and still be counted as Day 0 retained.
- Users who fire the starting event multiple times are still restricted to the calendar day they first fired the starting event. The exception to this is when a user fires the starting events on multiple calendar days; in this case, that user will be included in multiple interval cohorts.
Weekly and monthly retention using strict calendar dates
- Weekly and monthly calendar dates determine the beginning and end of each week and month.
- A week is defined based on the timezone specified in your project settings. There you’ll also find the option to specify the first day of your week.
- If your weeks start on Mondays and a user fires their first event on Monday, December 1st, Amplitude considers the user as Week 0 retained if they fire an event on any day between December 1st and December 7th. Amplitude will consider that user as Week 1 retained if they fire an event on any day between December 8th and December 14th. Another user who fires their first event on Tuesday, December 2nd has until December 7th to fire the return event in order to be Week 0 retained.
- A month is defined based on the timezone specified in your project settings. If a user fires their first event on December 3rd, Amplitude considers them as Month 0 retained if they fire the return event on any day between December 3rd and December 31st, and Month 1 retained if the user fires the return event on any day between January 1st and January 31st.
NOTE: For new user retention, filter conditions applied in the right module are only satisfied if they are true during the same time frame that the “new user” event was fired. For charts using strict calendar dates, this is the same as the chart interval. But for charts using unaligned ranges, the time frame is more granular: e.g. the first day for seven-day windows and the first hour for 24-hour windows.
Interpret your Retention Analysis chart: Retention view
Interpreting the Retention Analysis chart is more straightforward than it may at first appear, mostly because you can read through the parameters like a sentence. For example, the following chart shows you (1) new users who came back and did (2) any event, measured by (3) N-Day Retention (shown as retention and calculated by 24-hour windows) over (4) the last 45 days:
All these parameters can be easily changed to reflect the needs of your analysis.
In the rest of this section, we’ll explain the lower module, what all the parameter options mean, and how you can use them to generate the data you want. (The left and right modules are explained in our Help Center article on building your retention analysis.)
With the N-day retention line graph, you can easily see what percentage of users come back to fire the return event on a specific day. View the exact percentages by simply hovering over the data point for the day you’re interested in. If you want to inspect the users who make up that data point just click on it (see our Help Center article on Amplitude’s Microscope feature to learn more).
When you first open the Retention Analysis chart, the N-day retention graph will show the retention of new users who returned any event, until you add your own starting and return events in the left module (see our article on building a retention analysis).
If you’d prefer to see this data in bar graph form, click the icon. In this view, you’ll also be able to use Microscope to get more details on users who dropped off.
In addition to these graphic representations of your retention data, Amplitude provides you with a table displaying a detailed breakdown of the data, broken out by each user cohort and into more granular individual day buckets.
In this example, on January 29th there were 10,830 new users:
- The Day 1 retention is 22.9%, meaning that 2,480 of the 10,830 users came back and fired the return event between the 24th and 47th incremented hour after firing their starting event.
- The Day 3 retention is 21.6%, meaning that 2,339 of the 10,830 users came back and fired the return event between the 48th and 71st incremented hour after firing their starting event.
- Sometimes, there will be an asterisk next to the retention percentage. This means that the day has not finished yet and Amplitude is still collecting data for that particular day.
In the first row of the table (the "All users" segment), the value in the Users column is the total number of unique users who fired the starting event within the selected time frame. The user count for the present day will not be included until the following day, since Amplitude is still gathering that data.
While N-Day retention tells you the percentage of your users who come back and fire the return event on a specific day, Unbounded Retention tells you how many of your users fire that return event on a specific day or later. This helps you understand the overall churn patterns of your user base.
For example, if you want to know the percentage of users who have not used your product 30 days or more after first use, you’d use unbounded retention to find that out. The unbounded retention value for Day 30 tells you the percentage of users who returned no earlier than Day 30.
So, if a new user is inactive on Day 1, 2, and 3 but active on Day 4, the user will be counted as Day 1, 2, 3, and 4 retained in Unbounded Retention.
With unbounded retention, the tabular view works a little differently than it does for N-day retention:
- Assuming 10,830 new users on January 29th, the Day 1 retention is 36.6%, meaning that 3,963 of the 10,830 users fired the return event on Day 1 or later (at least 24 hours after the starting event).
- The Day 3 retention is 32.8%, meaning that 3,552 of the 10,830 users fired the return event on Day 3 or later (at least 72 hours after the starting event).
- As with N-day retention, results for days with incomplete data will have an asterisk.
Custom bracket retention
By default, Amplitude assumes you’ll want to use pre-defined units of time—days, weeks, months, etc—as retention brackets for your retention analyses. But you can change this, by using custom brackets instead of N-day or unbounded retention.
In the image above, we’ve defined four custom brackets:
- First Bracket: 1 day (Day 0)
- Second Bracket: 3 days (Day 1-3)
- Third Bracket: 3 days (Day 4-6)
- Fourth Bracket: 5 days (Day 7-11)
The line graph shows the weighted averages of all of the bracket retention numbers from the user cohorts within the selected timeframe.
In the table below, on January 29th there were 10,830 new users:
- The Day 1-3 retention is 35%, meaning that 3,791 of the 10,830 users fired the return event between the 24th and 95th incremented hour of their starting event.
- The Day 4-6 retention is 25.6%, meaning that 2,772 of the 10,830 users fired the return event between the 96th and 167th incremented hour of their starting event.
- Results for days with incomplete data will have an asterisk.
Retention vs change over time
Sometimes, you may need more than a straightforward view of your retention rates on specific days. You might want to know how a new release has affected your product’s Day 1 retention rates, or if a new training program has had an impact on your Day 14 retention rates. In these cases, you can view your retention data over time, by selecting Change Over Time from the shown as dropdown.
In this chart, we’re looking at all users who were new on November 29th. 25.5% of them fired the return event on Day 1, and 13.1% fired it on Day 7.
Amplitude calculates this percentage by dividing 1) the number of users from each new user cohort who fired the return event on each retention day, by 2) the number of users who were new on the selected day
The N-Day Change Over Time data table shows the same data as the N-Day Retention data table, but with the axes reversed (the X-axis is switched with the Y-axis).
Interpret your Retention Analysis chart: Usage Interval view
NOTE: Usage Interval View is only available with Scholarship, Growth, and Enterprise plans.
The usage interval shows the percentage of active users who’ve fired the selected event(s) with a specified daily, weekly, or monthly median frequency. Note that these event(s) must be fired on at least two different days in order for the users to be counted in the numerator. Your usage interval is important for drawing accurate conclusions about your retention numbers. Some products are built to be used daily, while others might be used much less frequently. Knowing how often your product is actually used will help you gauge the health of your product when looking at Retention Analysis and Lifecycle charts.
To view the usage interval, click Usage Interval View in the lower module.
Let's say your product's critical event is Play Song or Video. You can select this event and use the usage interval view to find the usage interval for that event. To calculate this, Amplitude plots the distribution of each user's median return period: For each user, Amplitude will look at all Play Song or Video events they fired in the last 15 days, and then determine the median length of time between each of these events.
For example, the highlighted data point tells us that 68.7% of your users have a median interval of 4 days or fewer between each Play Song or Video event. This inflection point can be interpreted as your usage interval. You can use this usage interval to create a custom bracket retention chart or a Lifecycle chart. In this case, it looks like 4 days is the expected usage interval for active users with the critical event of Play Song or Video.
To learn more about how to find your critical event, check out this blog post or our Product Analytics Playbook Vol. 1: Mastering Retention.
Usage Interval Over Time
When you select the Usage Interval Over Time view, you’ll see how the median frequency between events changes over time. Amplitude does not plot averages in this view; instead, it shows the actual percentages.
For example, the following data point shows us that of the users who fired Play Song or Video on December 8th, 88.3% of them fired it again within 7 days.