Customer Care Innovation

Analyzing Churn: From Prediction to Prevention

A novel end-to-end, data-driven churn analytics system that predicts customers who are likely to churn, identify drivers for churn, and the best ways to prevent churn.

From airlines to coffee shops to information services, loyalty and low churn rates have been the key to success. With open economy and easy access to alternatives in almost every sphere of lives of consumers, churn has become even bigger a problem for today’s companies. Market studies today report that industries such as banking (credit card) and entertainment (TV on Demand) suffer from 20% and 9% churn rate respectively, in the United States. Implications of churn can vary from one industry to another but a ballpark estimate says that a 30% churn rate leads to a negative RoI. Hence, every company in their industry needs to know their churn rates and implement means to reduce them, which will in turn increase the lifetime value of an audience, and ultimately, the profitability of the business model.

Our Approach

At Conduent Labs India, our researchers have come up with a state-of-the-art, novel, end-to-end, data-driven churn analytics technology. While traditional techniques can, at best, predict customers who are likely to churn, our technique can additionally identify drivers for churn as well as the best ways to prevent churn. Given consumer transactional data for a period of time, our technology would perform three kinds of analysis:

  • Prediction: A novel transductive technique for customer churn prediction using heat diffusion on hypergraphs.
  • Cause Identification: A maximal frequent pattern mining based approach for identifying interesting patterns of attributes of the churned customers, differentiating them from loyal customers.
  • Prevention: A novel influence maximization technique for modelling the spread of information in a customer network towards identifying a set of influential customers for intervention through incentives.

The system is implemented on a highly scalable Apache Spark platform where it can handle customer networks with millions of nodes and connections between them within tens of seconds.

Business Engagements

Ongoing pilot conversations with business groups in Commercial Healthcare and Financial Services.
Previous engagements with external clients including Groupon, Humana, KPN, Sprint, and Sunrise.

Technical Accomplishments

"Predicting the Next Customer Problem," Patent Filed.
"Customer Churn Prediction using Hyper-Graph Diffusion Analysis," Patent Filed.
"Predicting propensity for a customer to call and the preferred channel of communication based on historical data of customer interaction," Patent Filed.
"SentimentSIM: Sentiment aware Scalable Influence Maximization in Large Social Networks," Patent Filed.
"DIABLO: System and Methods for Root Cause Identification and Prediction Refinement of Customer Churn," Patent Filed.

Scientific Impact

"Debunking the Myths of Influence Maximization: An In-Depth Benchmarking Study," accepted for publication in the ACM Special Interest Group in Management Of Data (SIGMOD) Conference, 2017.
"Holistic Influence Maximization: Combining Scalability and Efficiency with Opinion-Aware Models," published in the ACM Special Interest Group in Management Of Data (SIGMOD) Conference, 2016.
"ASIM: A Scalable Algorithm for Influence Maximization under the Independent Cascade Model," published in the 24th International World Wide Web (WWW) Conference, 2015.