Customer Lifetime Value: Models, Metrics and a Multitude of Uses - Brian Bloniarz

[@zhaoxiaoxi0829]

"Poison do't churn" joke is awesome!

[@Corpsecreate]

Why are obvious drivers of lifetime value (like customer age) completely ignored?

[@brianbloniarz]

Great question -- the methods described here are about estimating future value of the customer base without covariates. One workaround is to segment the customer base and fit one model per segment. A smarter approach would be to learn the impact of covariates directly within the model. Doing that within these models is subtle mathematically -- two papers which do it are: "Incorporating Time-Invariant Covariates into the Pareto/NBD and BG/NBD Models" by Fader & Hardie, and "The Role of Time-Varying Contextual Factors in Latent Attrition Models for Customer Base Analysis" by Bachmann, Meierer & Naef. Both of those are implemented in CLVTools, which is an R package. Ideally models like these that allow for covariates would become more widespread.

[@Corpsecreate]

@@brianbloniarz I've modelled CLTV as a continuous state-space markov chain. A model learns E(Value) in year t and adds to that a bootstrapped estimate of LTV for year t+1. I use value iteration to get the estimates of each state to converge. The results are way better than anything shown here, and I think would be significantly better than the approach in those papers as well.

[@ravennsiregar]

@@Corpsecreate Hi Sam would you mind to give papers related to your approach by using continuous markov chain for CLTV?

[@Skydmig]

@@Corpsecreate do you have any article utilising this approach? I have been trying to find a good reference.

[@RidingWithGerdas]

Good presentation. Thank you.

[@MorrisonOscar-u6l]

Anderson Jessica Jackson Timothy Martinez Helen