Skip to main content

I really need help because I'm starting to pull my hair out on this supposedly "simple" subject. 

I would like to be able to analyze the retention at D1, D3, D7 and D30. 

For now, I just want to know which users are coming back after creating an account. 

To do this, I have set up a custom event called "Successful sign up" which, as the name suggests, is triggered when the user creates an account. 

(I can't leave the default "New user" event because my application is made so that you have to create an account to be considered as a user. It is possible to open the app and see the content without having an account. Therefore, if I leave it as is, I end up with a very low retention)

For the return event, I left the amplitude one, namely "Any active event". 

In fact:
- in D0 I have 100% with 115 users.
- in D1 I have 51% with 47 users.
- in D2 I have 43,3% with 29 users
- in D3 I have 31,9% with 15 users
- in D4 I have 24% with 6 users...
- in D5 I have 28,6% with 4 users ....

As you can see, all the numbers are wrong. I should have something like this instead:
- in D0 I have 100% with 115 users.
- in D1 I should have 40,9%.
- in D2 I should have 25%.
- in D3 I should have 13%
- in D4 I should have 5,2%
- in D5 I should have 3,5% 

 

To show you more the absurdity of these results, you can see on the screenshots below that my application gets a retention of 50% in D23 while there is only one user…
 

 

 

Hi Adrien,

 

Happy to help! I understand the frustration and our Retention chart has a couple of nuances that I can share to help. To begin, for your calculations that you did below:

As you can see, all the numbers are wrong. I should have something like this instead:
- in D0 I have 100% with 115 users.
- in D1 I should have 40,9%.
- in D2 I should have 25%.
- in D3 I should have 13%
- in D4 I should have 5,2%
- in D5 I should have 3,5% 

 

Those calculations were taken by dividing the number of users in each data point divided by total users correct? If that is the case that is not the formula we use for Retention. 

 

To begin, we only take users that have had time to convert so in Day 23 for example we do not include any users who have not had 23 days pass from their starting events. So, although there is only one user included right now there may be users whose 23 day to today who will not be included until the end of today. You can see users that are not included in the data point by looking at the breakdown table to see if there is an asterisk next to their value or not.

 

Now allow me to get into the calculation we use! The top row of percentages is the number of unique users who have performed the return event at Day X out of the total number of unique users who have completed Day X. In other words it is a weighted average percentage of the below row values.  You can read more on how each row is calculated here. The weighted average deduplicates users. So if users performs the first event and return multiple times they are only counted once per numerator and denominator.

Your Day 23 denominator is the deduplicated number of unique users who have had a full 23 days since their start event "Successful sign up". For your chart since you are looking at last 30 days on today May 31st it would be unique users who did their first event on Amplitude between May 1st and May 8th which is a small amount of users than your total of 115 users. The numerator would be the unique count of users that did the return event (Any Active event) on the 23rd day. 

 

I also recommend reading this other article I wrote detailing the calculation we use:

 

 

I hope this helps let me know if you have questions!

 

Best,

Sydney

Reply


Terms & ConditionsCookie settings