Finding drivers of long term retention

  • 16 March 2021
  • 2 replies

Userlevel 2


I’m working on finding drivers of long term retention in our SaaS product (long term in this case means that customers stick around for at least a year or more). I.e. the question I want to answer is: ‘What events or properties are leading indicators of long term retention?’

I’ve used both the Compass feature, as well as the cohort comparison feature to try to answer this question, but I find that it’s hard to use them with such a long time frame. 

I’m curious to hear how other Amplitude users would approach this problem!


Best answer by jarren.patao 18 March 2021, 00:21

View original

2 replies

Userlevel 3

Hi @morvik,

The Compass chart is a great feature for better understanding how your cohort performing a specific event correlates with them ending up in a specific user group (i.e. 4-week retention).

While we can’t provide a one-size-fits-all solution for driving long term retention, we do have our Retention Playbook which is a great resource to understanding your customer’s retention lifecycle as well as a post talking about this topic as well, but you can read about those here:

Some people look at what events are being performed at their highest rates to inspect user behavior around high activity (i.e. whereas others may actually look at areas that are showing lower metrics for retention so they can focus on driving growth in that area.

Another thing you could consider is inspecting users from longer retention buckets to see what actions they are performing to possibly promote more of those events which may be driving activity on your website or application (i.e.

Hope this helps! 

Userlevel 7
Badge +10

Hi @morvik ,
@jarren.patao brings up a lot of great approaches about exploring events/properties which can potentially indicate long-term retention.

Another approach I use every now and then is to look at the engagement matrix for my long term retained users and compare this with the users who drop out and see if I can find some events which are performed significantly more than the others.


Hope this helps!