Fighting Churn with Data

Churn is the bane of products and services with recurring users including subscriptions. Data Scientists, analysts and machine learning engineers are commonly called to the rescue in helping understand what causes users to churn, what is predictive of churn, and what can be done to reduce it. While examples of churn are available as benchmark data sets, analyzing churn in the real world is fraught with challenges and pitfalls for those without experience in this area. The first step is the hardest, and that's taking user information and a data warehouse of user product interactions and turning it into an analytic data set using sound feature engineering principles.

The second step is running some data science algorithms - that's the easy part! The final step is the most important: Turning the information produced into actions that can be taken by people in the business: marketing, sales, and customer facing support and success representatives. The training for most people only covers the second step (the easy part): running algorithms on a prepared benchmark data sets. This talk focuses on the crucial first and third steps: Feature engineering to create the dataset, and empowering different constituencies in the business to reduce churn with the results. The insights to be presented are based on the dozens of predictive churn analyses I have done for Zuora customers as Chief Data Scientist . The talk will features material from my forthcoming book: Fighting Churn with Data, to be published in 2019. (See www.fightchurnwithdata.com for more.)