Customer Churn Predictive Models

If you have repeat customers, then customer churn is a metric that you likely watch closely. Churn is a retention problem typified by the loss of repeat clients moving to your competition. The first step of resolving a customer churn problem is to get a clear understanding of why they are leaving. A better understanding of your their motivations will allow you can take steps to reduce the amount of churn. In some cases, businesses can prevent customers from leaving altogether. We use demographics and other profiling methods to create predictive models to identify those who are more likely to churn.

These predictive models will guide your ability to tailor your marketing strategies to better target customer that are in jeopardy of leaving with the messaging and solutions should increase the likelihood of them staying. Churn prevention costs are far less than acquisition costs of new clients, and because of this, the effectiveness of churn solutions are easily measurable when it concerns the return of investment (ROI).

Churn Data
To build a predictive model of churn, you start with a table of historical data about your customers. This table features columns to describe the customer that include customer demographics, such as age and location. Customer relationships change over time and these can help predict departures. To represent these changes, you can include columns that describe trends.

Lift Chart
A lift chart relates prediction confidence to the actual numbers of customers who churn. In the ideal case, the high confidence predictions correctly predict the majority of churned customers. Our client can directly target these customers for retention.

Decision Tree
Anova built a decision tree to predict customer churn. You can examine this tree to understand why customers churn, and directly use this information in your marketing strategy.

Validating Predictive Accuracy
The predictive accuracy of a model is the percentage of correct predictions. You can often increase prediction accuracy by adding additional data when building the model, and by experimenting with different types of models.

Important Predictors
You can use the predictive values found here to understand the reasons that customers depart. For example, in the demo data set, the previous year’s transaction count and region code are important predictors, while age and gender do not seem to affect churn rate.

Churn Prediction to Actionable Insight
Finally, you can see how the two most important features relate to the confidence in predicting churn. These findings suggest that customers in a particular geographic region are poorly served, and can be targeted for an improved relationship.

Anova Analytics is a nimble Data Science boutique focused on standing up and team empowerment through skills, methodology, and technology training. We combine this training with consulting and infrastructure solutions for our clients. Anova Analytics offers an online platform which provides the methodology for your team to build and grow together. While working with your team, we can help with your project to provide you with tangible and relevant models that work.