The company is an established provider with over 70,000 subscribers. With a 15% annual churn rate, they needed a way to protect revenue by keeping existing customers rather than just focusing on new sign-ups.
The main challenge was timing. Reaching out during a cancellation call is usually too late because the customer has often already signed a contract with a competitor. At this stage, trust is lost and even deep discounts rarely work. The company needed to identify unhappy customers while they were still in a position where they could be saved.
The project began by gathering all customer information into one place. This included payment history and technical logs, support calls etc. We connected data that was previously stored in different departments. This provided a unified view of how each subscriber experienced the service over time.
We used the organized data to find specific behaviors that happened before a customer decided to quit. These signals acted as quiet clues. Examples included a slight drop in internet speed or a sudden change in how a customer paid their bills. These small signals were often too subtle for people to notice manually.
We trained a predictive model using thousands of examples of past customers who had already left. The model learned to recognize the patterns that led to those departures. The system now assigns a risk score to every subscriber based on these patterns. This allowed the provider to predict an exit before a formal complaint was ever made.
The retention team began receiving a weekly list of these at-risk customers. Instead of waiting for a complaint, they called the customer with a direct solution. They fixed technical glitches or offered better price plans while the customer was still happy to talk. This approach turned a potential cancellation into a positive service experience.
