Using Predictive Analytics to Pinch the Client Leak

This past week was an exciting one for folks in the machine-learning/predictive modelling world.  No, it was not Albert Einstein’s birthday (March 14) nor was an  amazing invention by Nikola Tesla suddenly discovered. While any self-respecting data scientist, and those who work with said scientists, would be excited about either of those two events, something more subtle occurred: a book was published that helps translate the amazing world of predictive modelling into a language that even a knuckle-dragger like me can understand. It is called Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die. It is written by Eric Siegel and I highly recommend you buy this work.

Churn Modeling in Predictive Analytics

One of the chapters describes a technique that our company has deployed for some of our clients and that is using predictive modeling to identify a client who may be at a higher risk of leaving to another vendor. This application of predictive modeling is described as “churn modeling” and in the book it is likened to a balloon with two holes: one representing new clients entering the balloon to create a company’s client base. The other hole in the balloon represents lost customers who have left for another vendor.

The method to churn modeling is to take large amounts of past data to include: client ordering patterns, part numbers, cancellation history, payment patterns, complaints/feedback, call history and any other remotely applicable information surrounding a company’s customer base, and apply machine learning to the data. The machine learning techniques will identify trends and momentum characteristics common amongst the customers who were lost as well as those who were retained. This information then can be translated into a predictive model, or algorithm, against which all existing and future customers can be “scored” for likelihood to be lost.

Simple-Balloon-NLP-Graphic-2 (ID 25235)

Predictive Analytics Stop Leaks

The great thing about this approach is the model will only get smarter as you use it more and more. New attributes will appear and old ones may drop off as market conditions change for both you and your customers but you are ready. You’ve optimized your data to give you a look into the future and to begin to squeeze that customer leak.
Of course, now that you know which client may be leaving…what are you going to do about it. Yes, there is a predictive model for that too but I will leave that for a future post.

About the Author:

Ted_LinkedIn (ID 25955) Ted Willich is a team-building senior leader with operations, business development, strategic account sales and start-up experience.  He is the CEO and Co-Founder of NLP Logix, Predictive Modeling and Machine Learning as a Service.

0 Pings & Trackbacks

Leave a Reply