1
$\begingroup$

I'm trying to work out some explanation for a result I'm seeing where I have the Kaplan-Meier plots for monthly customer churn risk. In aggregate, it looks fine, but I've broken it down into subgroups based on whether the user has previously churned before, i.e. canceled and resubscribed.

A couple things that might be impacting this -- I'm looking at a shorter time horizon than I'd like (~3 years relative to maybe about 10 years of data). More customers are likely to have not ever churned before, so the other curves represent smaller numbers in total, but still >50k.

Leaving aside the fact that recurrent event analysis is probably the better approach if I had access to the full data, I'd love any insights if anyone's seen something like this before. It does also appear in the duration distribution comparison, with non-churners having a very different distribution.

My main hypothesis to explain this is that this plot suggests that customers will, more or less, inevitably churn, so customers who have already done so in the past are less at risk now. Similarly that customers who give the service a second (or third, etc.) chance are more likely to stick with it?

And then there's the idea that a user who has already churned >= 1 time will probabilistically, in aggregate, have shorter durations -- given this, I'm wondering how I can correct for this effect?

It's not a particularly satisfying answer to me so far. I just wish I had a more quantitative way of demonstrating this.

KM curves for users, lower survival rate if user has never churned before

Distribution of durations

$\endgroup$
2
  • $\begingroup$ "My main hypothesis to explain this is " explain what? "anyone's seen something like this before" seen what? $\endgroup$ Commented Mar 30, 2023 at 16:36
  • $\begingroup$ If this question is about 'why do customers churn less fast after each time that they have churned before?' then it is about some domain specific theories and not about statistics. $\endgroup$ Commented Mar 30, 2023 at 16:48

1 Answer 1

2
$\begingroup$

This is not counterintuitive at all.

Customers who churned before and resubscribed had to have a reason to do so. The range of reasons depends on your service but a typical one is that they tried a competitor and decided that your service is better after all.

Resubscribers have more conviction about your service because they have more experience with the landscape.

Don't confuse the conditional probability of staying longer given resubscription with the probability of resubscription.

Do you have any data about why your customers churn? We ask them on the cancellation form.

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.