Cloud Hosting

I have learned three things over time: 1) Just because an excuse is correct doesnt make it any less of an excuse 2) More value is added to a discussion by asking the right questions than by giving the right answers 3) You cannot say half of the job is done in stand-up comedy by just standing up. I am a little carried away by the second one today. I am going to discuss the topic of churn analytics by only asking questions. You tell me whether the questions guide you in the right direction.

How do we define a churn so that we identify exactly how many and when churns have happened in the past? Is it the event when the status of the customer was updated as inactive in the IT system? Would defining churn with a reduction in product or service usage by more than a threshold number be more helpful? Is it when the customer didnt make his bill payments for several consecutive months? What signals or patterns are more likely to occur in case of customers about to churn? How do patterns differ among customers who are not at risk of churn? Should we consider churn a binary event (churn, no churn) or a multi-level event with various churn gradations?

How much in advance should we predict churn? How much lead time does the business team need to act on at-risk customers? What offers or campaigns suit a customer predicted to churn in 3 months?

What urgent incentives do you offer a customer predicted to churn next week? Is it ok to let specific customers churn? Should we consider a customers lifetime value (LTV) prediction to decide how much criticality to assign to save the customer? How do we calculate the effectiveness of a churn prediction? If the customer is predicted to churn but doesnt, is it a good churn prevention action or a bad prediction? Isnt it always a wrong prediction if the customer is not predicted to churn but churns? Dont you think because the customer was considered to be not at risk, no action was taken?

How does a customers journey look till the point of churn? Does that give us a good peep into the factors causing churn? Was the outcome of a customer touch point with the organisation negative? How many touch points went negative for a customer? Which channels (automated IVR, call centre, mobile app, website, stores, in-person interaction, SMS, email, WhatsApp) mattered more in influencing churn? On average, have we taken more time to solve tickets for customers who eventually churned compared to those who didnt? Were quantitative responses to customer survey questions harsher from customers who churned? Was the sentiment derived from a qualitative response negative from a customer who churned? Does churn increase with decreasing NPS and vice versa?

Which product and service areas have a higher churn rate? How does the tenure of customers correlate with the churn rate? Are new customers more likely to churn compared to older ones? Customers who have a higher frequency of touch points churn more or less? Did the customer have a good experience during onboarding? What does the trend of churn rate look like? Is there any seasonality? Is there any other pattern in the movement? How many times has the customer defaulted on bill payments?

Should we design the churn problem as a binary classification problem or a multi-class classification problem? Before building a churn prediction model, should we cluster the customers first into different groups through an unsupervised learning technique? Which influencers of churn can be controlled (for example, customer experience)? Which influencers of churn are outside the organisations control (for example, economic slowdown)? When I say something impacts churn negatively, isnt it ambiguous? Instead, shouldnt I say something decreases the churn rate or something reduces the churn numbers? Shouldnt I be clear on whether new customer additions offset churn numbers? Or is churn calculation independent of how many new customers the organisation has added? Are churn numbers and churn rates calculated on a monthly frequency? Half-yearly? Annually? What is the target reduction in churn rate through this prediction solution? Do we know how much monetary savings it translates to if we reduce churn by the target rate?

Views expressed above are the author's own.

END OF ARTICLE

See the rest here:

For whoever thought data scientists knew churn prediction - Times of India