Churn is known by other names such as customer attrition and customer turnover.  It is a well-recognized problem in the telecom industry.  Service providers go to great lengths to prevent or at least reduce the voluntary churn-out because they recognize that retaining an existing customer is much easier than acquiring a new one. In the telecommunication industry consumers are often tempted to switch to the first provider who is willing to pay them out of their existing contractual agreement or offer a free upgrade. As a result, the vertical suffers from a high churn rate.

According to Gartner the telecommunication industry’s market size is estimated to grow 89% YoY to ~ $4.2 billion with increasing 5G infrastructure and applications [Delloite]. There are a number of solutions in the market that assess the customer's propensity to churn.  The telecom industry is currently experiencing a churn rate of nearly 38% globally.  Active poaching and strides in technology and technology-adoption are key contributors to this constant exodus of customers from one provider to another. Advanced prediction can help design personalized offerings to reduce churn propensity and improve profits.

Sparse Weighted Auto-encoder or SWAN is a novel approach to the problem.  The use of multi-dimensional modeling based on mobile-app usage as opposed to traditional voice call consumption metrics to predict when customers will churn is one of the key differentiators of SWAN.  ChrysalisGold has demonstrated churn predictions with accuracies reaching 85% with this approach.