Client churn approach

Hi there, I’m starting to do the EDA process with the objective of creating a Machine Learning model that predicts the likelihood of client churn for the company where I work. We have the relation that 1 account can have multiple opportunities (services), so what do you think is the best approach to measure the likelihood of the churn, taking in consideration churn just until an account has canceled all the services OR considering churn with each service?

The reason I’m asking this is that the decision to churn is often based on a combination of factors, including the overall satisfaction with the company and the relative value of the services being offered. By considering the churn of the entire account, you can take these factors into account and potentially improve the performance of your model. But I want a second opinion on this.

What do you think?

Thank you so much in advance.

I’d say overall there’s not enough information here to really provide an informed opinion, but I know that time since activity can be a strong indicator of churn. Really depends on the industry and data availability. Aim for feature creation?

Edit:
rereading this, I think it is likely that you will find with the churn of one service there may be a cascade effect into other services like you’re hypothesizing. If you aim at predicting overall churn then it is likely that churn in a sub-service will be a strong predictor for overall churn. Answer is still the same though, aim at feature creation for predicting sub-service churn and overall churn with sub-service churn probability being a feature in overall churn. Hope I’m understanding this right! :+1:

2 Likes