Knowing who’s leaving before they do
A regional telco had retention offers but no clear idea who needed them. They wanted to act before the port request.
A regional telco had retention offers but no clear idea who needed them. They wanted to act before the port request.
Churn was reactive. By the time a customer called to cancel or port out, it was often too late. Marketing had segments and campaigns but no model that predicted which subscribers were at risk in the next 30–90 days. They needed a score and a recommended action so care and retention could focus where it mattered.
We built a churn prediction and next-best-action layer on top of their billing and usage data. It scores each subscriber for likelihood to churn and, where possible, suggests a retention offer (discount, plan change, perk) that fits their profile and constraints. Scores and recommendations are refreshed regularly and pushed to their CRM and care tools so agents see the signal when they’re on a call or when a high-risk customer hits the website. We trained on historical churn and made sure the model was interpretable enough for marketing and ops to trust it.
Retention campaigns became targeted instead of spray-and-pray. Save rates on at-risk customers improved, and the cost per save dropped because they weren’t over-offering to stable subscribers. The telco is now using the same scores to prioritize network and service improvements in high-churn areas.