How can I improve the CLV accuracy from lifetimes model?

Question

Not sure whether this should be asked in StackOverflow or here, but I believe this question is most about statistics, so I'll try it here.

So I was modelling the Customer Lifetime Value using lifetimes package. However, the overall accuracy is not so good. There isn't much documentation about the model, except for the one provided in the official website (this one: https://lifetimes.readthedocs.io/en/latest/).

Can anyone help me with this? Any ideas on how to improve this model accuracy?

Related-if-not-duplicate: How to know that your machine learning problem is hopeless? — Stephan Kolassa, Nov 03 '20 at 20:05
While the Python package lifetimes does not offer this possiblity, you can include covariates when estimating CLV with the R package CLVTools (https://github.com/bachmannpatrick/CLVTools). Here is an exemplary analysis for a fashion retailer, which illustrates how to perfom such an analysis: https://www.clvtools.com/articles/CLVTools.html — majom, Nov 19 '20 at 17:34

score 0 · Answer 1 · answered Nov 20 '21 at 15:15

In the context of mobile gaming, this paper showed some interesting properties about CLV that can be used to improve modeling.

For instance, it suggests that one could be benefit from splitting the modeling problem in two parts:

[Conversion] Model the CLV for users who haven't converted yet (i.e., users with zero purchases).
[Long-term monetization] Model the CLV for users who have paid at least once before.

This observation is based on the analysis of time in between purchases in free-to-play mobile games. But arguably a similar principle should apply to any product with a freemium business model.

How can I improve the CLV accuracy from lifetimes model?

1 Answers1