How the Retention Analysis chart calculates retention

Amplitude's methods for calculating retention are direct, but you should familiarize yourself with the differences between them. This helps you develop a nuanced understanding of the Retention Analysis chart.

Return On or After

Amplitude calculates Return On or After retention a bit differently depending on whether you're looking at a specific cohort entry date for your segment or overall. Overall retention (for example, All Users) appears in the visualization and the first row of the breakdown table below it.

Because Return On or After retention measures users that returned on the Xth day or later, the numerator includes users in all data points prior to when they triggered an event. A user who triggers the event on day two, for example, also appears in the data point for days one and zero.

Specific cohort entry date

The calculation for a specific cohort entry date is:

{ # of users who triggered return event on X day after cohort entry date,
or any day after X day }

divided by

{# of users who triggered start event on specific cohort entry date
[constant across X days] }

In the table below, Day 3 retention for April 1 shows that 6,633 users triggered the return event on April 4 or later. 9,644 users triggered the start event on April 1. Dividing 6,633 by 9,644 gives an April 1 Day 3 retention of 68.8%.

The retention percentage for a specific cohort entry date always decreases over time. While the denominator remains constant, the numerator gradually goes down: the number of people who are 7 day+ retained is less than the number of people who are 3 day+ retained.

Overall retention

The calculation for overall retention (also known as all users) is:

{ # of users who triggered return event on X day or later after cohort entry date }

divided by

{ # of users who triggered start event on cohort entry dates
that reached the retention interval }

In the table below, the All Users Day 3 retention shows that 55,752 users triggered the return event on Day 3 or later, while 137,586 users triggered the start event between March 30 and April 6. Dividing 55,752 by 137,586 gives a Day 3 retention of 40%.

In the table, the All Users row contains the daily sums of the date rows below it. Amplitude excludes incomplete data from the All Users totals (incomplete cells have an asterisk).

When the analysis time frame is complete, the graph curves down as retention decreases over time. This happens because fewer people can be retained for longer periods of time as shown in the numerator (it's harder to retain people for seven days than for three days).

However, when the analysis is still in progress, the graph can curve up and appear to increase over time. This happens because Amplitude excludes users who haven't yet reached later retention intervals from the denominator. This means Amplitude hasn't collected enough data, and you should give your users more time to trigger your return event. Wait until your analysis time frame is fully complete to get an accurate understanding of your retention trend.

Accordingly, the denominator value in the Microscope for a single day is the sum of the users who completed that day's retention interval. The breakdown table doesn't show this value.

Return On

The calculation for Return On retention is:

{ # of users who triggered return event on exactly X days after cohort entry date }

divided by

{ # of users who triggered start event on specific cohort entry date
[constant across X days] }

In the table below, Day 3 retention for April 1 shows that 6,594 users triggered the return event on April 4, while 9,644 users triggered the start event on April 1. Dividing 6,594 by 9,644 gives an April 1 Day 3 retention of 68.4%.

All users' Day 3 retention shows 48,219 users triggered the return event on Day 3, while 125,665 users triggered the start event between March 30 and April 6, giving a Day 3 retention of 72%.

A user can trigger the return event on multiple days and Amplitude counts them on each day. This can drive retention percentage up over time, for a specific cohort entry date row or overall row. While the denominator is constant, the number of users that trigger a return event each X days is independent. More users can trigger an event on any day than the prior.

As with Return On or After retention, when the analysis time frame is complete, the denominator value in the Microscope for any day is consistent with the total number of users for overall retention. When the analysis is still in progress, the denominator value in the Microscope for a single day is the sum of the users who completed that day's retention interval. The breakdown table doesn't show this value.

The overall row represents the sum of the cohort entry date rows below it. Similar to Return On or After retention, if data is incomplete (cells noted with an *), Amplitude excludes them from the overall row's total (for example, the sum of each row for Day 3 doesn't add to the overall Day 3 value).