- Rather than go for churn by cohort or focus on churn I think it's better to create a model to predict user LTV. I think this is hugely valuable and the graphs reflect a lot better as well when rooted in LTV rather than churn. - At the end of the day they both correlate but given we're offering methods to prevent churn and to ask people to talk to users likely to churn this seems better. This also feed well in to ranking users in both [[Product Feedback]] 4th graph as well as [[User Churn]] 4th graph. ### Approach - Figure out how to model LTV. Have to find some formula to capture current LTV of users and then divide into training and testing. Basically building a regression model that predicts the LTV of users. We can allow the users to pick properties but the most seamless way would be for us to do it ourselves either through ML or with some manual input. - Graph 1 - We show total average, top 25% , top 10% and top 1% LTV values by source - Graph 2 - Features of users with highest LTV. These can be characteristics of users. For example how phone price ended up being a strong indicator for WorkNetwork. - Graph 3 - We show the list of highest LTV users on the application as well as their associated LTV - Graph 4 - This may be hard to do but in the long run we can show actions that drastically increase LTV of a user. Like what actions they would take that might increase their LTV. I'm not sure how this would emerge from the model but something that would be very valuable if we can pull it off.