Customer churn has been a growing problem across industries. Businesses are losing millions of dollars every year to cancellations, and it is affecting their balance sheets.
In recent years, the advent of predictive analytics and its initial success in retaining customers has given hope to leaders of these companies. For example, video streaming giant Netflix recently claimed to have saved almost $1 billion by retaining customers using predictive algorithms. Successes like this have resulted in the rising popularity of predictive customer churn models over the last decade.
In such an environment, many retention leaders are focusing on developing churn prediction models with a much-focused agenda of identifying which of their customers are more likely to cancel. With so much visibility and attention, a data scientist would work on just one churn model with a focus on improving its value by measuring how well the outcome is predicted and in customer retention scenario, it would be how many and how accurately the model has identified at-risk customers.
Why Customer Churn Still Isn’t Slowing Down
In 2018, the churn rate in most industries across the U.S. was greater than 20%. Why are companies still struggling to contain churn despite putting in place an advanced churn prediction model and investing in retention programs?
The reason lies in how machine learning models work. It is important to understand that a machine learning model can answer only one question at a time. Most predictive churn models are effectively focused solely on producing a more accurate and refined classification of customers between “at risk” and “not at risk.” Apparently, this classification alone is not sufficient to actually reduce churn.
Why A Single Churn Model Is Ineffective
1. No insight into the context of risk: The company may be aware of who its high-risk customers are, but it won’t know from the model output why a particular customer would want to cancel in the first place.
2. No clarity about customer value: A churn prediction model doesn’t tell you which of the identified at-risk customers is more valuable. In this scenario, your retention agents end up giving costly offers to low-value customers.
3. It doesn’t allow timely and proactive engagement: Marketers do not know how much time they have before a predicted high-risk customer will cancel. This will hinder their planning on who to target first.
4. The lost opportunity of customer winback: Winning back lost customers is more profitable than new customer acquisition. The single model approach will only predict the risk status of active customers and won’t even consider winback chances of recently canceled customers.
Multiple Predictive Models Approach For Customer Retention Boost