Creating New Users Cohort

  • 11 December 2019
  • 7 replies


Following the post about Lifecycle analysis, how do I create a New Users cohort?

@Marco from Mixpanel Education - maybe you can share screenshots for all the cohorts you mentioned in the post?


Best answer by marco 24 January 2020, 18:51

View original

7 replies

Userlevel 1

Hey @sharod,


Here's an example of the cohort builder for a new user cohort. Here are the assumptions for this example:

1. The key event that matters to me for this product is "Watch Video".

2. The usage interval that aligns with this product is 7 days.

Based on that, I define new users a someone who performed "Watch Video" in the last 7 days (that's A in the screenshot below), but did NOT perform "Post video" between 14 and 7 days ago (B), or 21 to 14 days ago (C).

The whole point of B and C is just to make sure that this is actually a new user. So depending on your use case, you B and C could be longer. The main thing would be to avoid accidentally counting a user that was active a long time ago and then came back, cause you'd wanna put that user into a resurrected cohort, not a new user one.


For my product here, I don't think I have any users who were around before April 23rd of this year, so I don't have to worry about that.


Here are some screenshots of how I would define the other cohorts with the same key event and usage interval:


RETAINED (perform event in two consecutive intervals)


RESURRECTED (performed event, then didn't, then did it again in the most recent interval)

DORMANT (performed key event in previous usage interval, but did not in the current one)

And notice how I did rolling range with all the dates so that it's regularly looking for users who meet that criteria!


hope this helps! —

@marco Hey Marko. This elaborates the static aspect of New/Returning/Dormant user cohorts building process but what if I want to look at the trend of these users over time.

Is there a solve for that? I want to set a report which shows me day over day trend of new users coming on to my platform. Does the cohort adjust accordingly?

Userlevel 4
Badge +4

Hi @kashyapbarua: when you take these cohorts with the relative date ranges (last 7 days, between 14 and 7 days ago..) and visualize the trends over time, we dynamically compute the cohort size as of each of those data points.


New users cohort definition:


Visualized cohort trend over time (weekly breakdown):

Hope this helps? Let me know if you have any other questions.



hey @marco you mentioned “I don't think I have any users who were around before April 23rd of this year,” - how would you setup this analysis if that wasn’t the case?


Also a slightly separate question - how generally do you go about separating your weekly active user cohorts and monthly active user cohorts if the only thing that seperates them is their key usage metric? 

Hi @marco thank you so much for this post, super helpful! 

I have a few questions:

  • Is it ok to apply the “1sr time ever” filter in filter “A” for new users to make sure they aren’t “Resurrected” ones?
  • For Resurrected & Dormant users, do we want to include those who first appeared on the product 3 weeks ago (filter C)? Or it doesn’t matter for this use case if it was there first time back then?

Thank you!


Userlevel 4
Badge +4

Hi @Maria K , Moinak here from the product team.

“1st time filter” is perfect to define new users who did event A for the very first time in the last X days.

For resurrected, I would do something like this (the layout of the QB might look a bit different from yours, but this newest update is coming to your reports in the next few days!):

For retained, I’d use something like this:

For dormant, I’d use something like this:

Hope this helps? Let me know if you have any other questions.

Hi @moinakb !

Really appreciate your detailed explanation on the cohort filtering, which I did replicate for my own analysis. But there’s missing users / wrongly grouped users when I compare each category cohorts (made for last 30 days period) with calendar month period data.

A (Event uniques) vs B+C+D (Lifecycle Cohorts)

Do you have any tips/suggestion to exactly create/group these cohorts according to the calendar months?