Insights

Churn Prediction: Five Success Stories

Yogesh Sharma

> Download

Firms of every size and industry face challenges of churn among customers, employees and sales agents. The cost of acquiring customers is several times that of retaining a customer, which is why firms continually search for solutions to this pervasive issue. With the availability of data and advancements in data science, predicting churn is getting much more sophisticated. The ability to predict churn and, more importantly, design appropriate intervention strategies at the subject level (customer, agent and employees) is key to controlling the associated costs.

Churn is triggered by several stimuli (performance of product or service, client issues, competitive and technology landscape). With the advances in technology and data science it is becoming possible to gather much richer and larger data on these stimuli and apply data science to predict the churn and identify the drivers. However, across different firms we have observed that maturity level of infrastructure and capabilities to gather the data and develop data driven churn prediction process varies.. However, similar effort in developing a data driven churn management meets with several challenges. These challenges are in the form of:

  • Recognition and definition of churn: How easily is churn defined and recognized to develop prediction to mitigate it. Slow revenue leakage from an account is less visible churn than a customer closing the account.
  • Data sourcing and data science: When churn is not easily defined and recognized, effort to secure data helpful in predicting churn often suffer, resulting in no or sub-optimal models to predict churn. Various data gaps and associated IT infrastructure do not get adequate management focus. Even with available data, skills and experience required to deploy sophisticated analytics models and machine learning algorithms vary.
  • Change management: Data driven recommendations have to compete with historical and gut based insights and decisions by stakeholders (account managers, HR, sales force) forcing the need for a programmatic change management so that algorithmic recommendations are adopted by relevant managers.
  • Operationalizing the churn management. Enabling the actions and impact measurement and generate real-time signals is essential for managers to take action on the identified drivers and prevent loss of revenue. In the absence of such system, the role of data science and modeling remains a one-time effort.

The intensity of above challenges vary across different firms and functions and also, is characterized by types of industries. Elements of a good churn management program include sound churn definition, an inventory of available data to develop a churn-prediction model, identify additional data gaps, running a proof of concept model and evaluate the quantified result to stakeholders, and develop a full-scale reporting and feedback system for the stakeholders. The degree to which these different elements are required would be determined by the stage at which a firm is along above outlined challenges. In the attached paper, we present our experience in working with firms in different industries to developing data driven churn prediction capabilities.


About the Experts





Yogesh Sharma is an associate principal in ZS’s New Delhi office and is a leader in the firm’s advanced data science practice. He helps clients develop data science capabilities in the marketing arena, including customer acquisition, life cycle management, and insights driven decisions. Yogesh has more than 20 years of experience in customer analytics across financial services, retail, utilities, airlines, casino and hospitality, telecom and big data. He has helped clients across these industries to leverage enterprise transactional data for making better business decisions.