Hi,

We have defined the following:

Power User = fires event A on 5 or more distinct days of the week

Core User = fires event A between 1 and 4 distinct days of the week 

Casual User = fires event A at least once a month, but not frequently enough to be a Core or Power user

Dormant User = Has gone more than 30 days without firing event A

I am trying to create a chart that has weekly cadence, and shows me how my user population breaks down into these categories over time. Something like below (incorrect definitions used for the screenshot, but just wanted to convey what I envision the final product looking like). 

Week to week, if a user’s behavior changes, they would be represented in a different category. Which category a user is in for a given week depends only on the behavior they showed during that week (or during the last 30 days relative to that week if we’re talking about Dormant category).

I tried using Behavioral Cohorts first, since I know there is an option to use “count in interval” and require that a user completes an action at least X times within a given interval, e.g. week. The problem with this, is that the “cohort population over time” feature is not supported for this kind of cohort, and if I try to use the cohort in a chart, it will just show me behavior from the users CURRENTLY in the cohort, and not from the users that may have fit the cohort definition at some point in the past.

I have also tried in-line cohorts in the event segmentation chart like “users who did event A >= X times any time in each week”, but the issue with this is that it can only be based on the number of events a user fired, and not on the number of distinct days during a week that they fired the event.

Has anyone figured out a way to achieve my desired result? Its kind of a marriage between some of the functionality the cohorts allow + some of the functionality the event segmentation user segment panels allow… but I haven’t figured it out yet. 

I have read through numerous blog posts from Amplitude and can’t find something relevant, since my use case seems to be unique in that we are defining these states based on the number of days in a week an event fires.

Thanks in advance!