Methods for Modelling Customer Lifetime Value: The Good Stuff and the Gotchas

Part three of a comprehensive, practical guide to CLV techniques and real-world use-cases

How often does a customer shop? How much do they spend? And how long are they loyal? Three simple factors to help you model your average consumer’s Customer Lifetime Value. But does that make it an easy task? No. No it does not. Source: Author provided.

Welcome back to my series on Customer Lifetime Value Prediction, which I’m calling, “All the stuff the other tutorials left out.” In part one, I covered the oft-under-appreciated stage of historic CLV analysis, and what you can already do with such rearwards-looking information. Next, I presented a tonne of use-cases for CLV prediction, going way further than the typically limited examples I’ve seen in other posts on this topic. Now, it’s time for the practical part, including everything my data science team and I learned while working with real-world data and customers.

Once again, there’s just too much juicy information for me to fit into one blog post, without turning it into an Odyssey. So today I’ll focus on modelling historic CLV, which, as part one showed, can already be very useful. I’ll cover the Stupid Simple Formula, Cohort Analysis, and RFM approaches, including the pros and cons I discovered for each. Next time I’ll do the same but for CLV prediction methods. And I’ll finish the whole series with a data scientists’ learned best practices on how to do CLV right.