Cox regression seems useful as a broad strokes model for measuring survival/ churn of a subscription-based product/ account lifetime estimate -- and the broad factors that might influence the survival curve (and area under the curve for estimated lifespan).
Unlike typical human survival data however where there's generally a live/ die binary ---
In business terms, unlike with an account "dying" -- the living accounts can "grow" in terms of monthly/ annual revenue.
In other words, if one were building a model to predict "lifetime value" (estimated lifespan x (value per t ) .... would one combine a linear regression model to estimate "growth/ value" over time ... to combine with 'estimated lifespan remaining' essentially?
This seems like a fairly common business problem. I've been playing around with a few things -- and the Cox portion seems straightforward. In terms of estimating lifespan based on a few key variables & current age --
The other part seems complicated. Like sure, you can do a very basic 'any account that survives grows 10% per month" -- but in reality there are many measurable variables that would impact this prediction as well.
I suppose the models wouldn't be combined per se.... there would just be a Cox model for lifespan ... a regression model (or some sort of time series) -- for estimated 'Revenue per T' ... and then essentially a cross-product of some sort of the outputs...
