Funnel problem: merged users are counted double in the initial audience

  • 3 December 2020
  • 1 reply

Hello, we have an odd problem with the funnel.

This is the funnel: app install → get_started button →  login →  first onboarding question.


If I create a cohort with the get_started drop-off, in this way:


I find users with the login event, listed in this cohort!


In the same cohort, I find also the anonymous user and the identified user (merged with the anonymous user). In this way, the initial audience is counted twice, and the funnel percentage is wrong.

If I add the filter: UserId is set, then I have the right results and the drop-off cohort is populated with the right users (no login events).


It seems like the funnel includes in the initial audience the anonymous users and the identified users also if they have been merged (auto-merge is active in our account).

Note: we are using segment to send the events and to identify the users.


Best answer by cherise 17 December 2020, 19:54

View original

1 reply

Userlevel 6
Badge +3

@simone.c I can see that you are working with our support team on this, but wanted to share some initial steps here: 


  1. One thought is that if the conversion window is short, like a day, it would make sense to see users who did Login in your step 2 dropoff cohort, if that login event took place more than a day after the step 1 event. 
  2. Secondly, that user_id is set filter that you have globally makes sense in terms of how it changes your funnel numbers. The user ID property only gets set once the user has been .identified, which typically happens when they log in or sign up - meaning that users who never did step 3 will end up getting filtered out of the funnel because their first two events do not include this property. 
  3. One other thing to note - with ID Merge, sometimes reports can take a few hours to reflect merged data, so for the newest users, they might count as two separate users even if they've been merged, if they were merged just a couple of hours ago.