How do you interpret churn analysis?

Data analysis results are as powerful as their outcome. The more interpretable the results show greater business impact. In this post, I am aiming to create a churn analysis and monitoring dashboard. This dashboard helps us ;

- Identification about who is churning?

- Comparing churners and non-churners at the different type of features

- Automated analysis will save time you don’t have to do all over again and again

I will use a Kaggle data set, churn for financial services. First I will create a dashboard to see overall at a glance and use Tableau than I will calculate correlation figures between variables with python. Churned is the target variable and the others are independent variables. Data set includes:

1- Customer id

2- Geography — a customer’s location can affect their decision to leave the bank also countries have different churn rate.

3- Tenure — refers to the number of years that the customer has been a client of the bank.

4- Balance — also a very good indicator of customer churn, as people with a higher balance in their accounts are less likely to leave the bank compared to those with lower balances.

5- NumOfProducts — refers to the number of products that a customer has purchased through the bank.

6- HasCrCard — denotes whether or not a customer has a credit card. This column is also relevant, since people with a credit card are less likely to leave the bank.

7- IsActiveMember — active customers are less likely to leave the bank.

8- EstimatedSalary — as with balance, people with lower salaries are more likely to leave the bank compared to those with higher salaries.

9- Exited(Churned) — whether or not the customer left the bank.

Identify your churning customers and their pattern.

To frame the dataset/customers I choose sharing the major numbers on the top of the dashboard, this will help the reader to follow the answers to these questions How many customers we have, How many of them are leaving? What are their general figures?

You may also see on Tableau public

Country churn rate: Bank’s overall churn rate is 80%, France has 84%, Germany 68%, Spain 83% churner customer. Spain and Germany have a similar count of customers however Germany has a lower churn ratio. Comparing Spain and Germany might spot good do’s and don’ts.

Customer tenure: Working with the same bank for long years does not change the decision to leave the bank. Tenure vs Exited graph shows that whether tenure years are less than five years or more than five, they still have a similar churn ratio.

Product Usage Effects on Churn: Average number of products per customer 1.54 for churners and 1.48 for non-churners. Question: why an average customer does not have more than two products?

This might be indicating a cross-selling ineffectiveness.

Credit Card Usage: A customer has only one product on average, usually, this only product is the credit card. Thus, deep dive into card usage with dividing two parts; first credit card user, secondly credit card user but also has more than 1 product.

Churners and non-churners have similar ratios at the same dimensions (cc usage), this flag still does not explain churn.

Age and Balance: Age has right-skewed distribution and the mean is 38. Between 45 and 64 non-churner ratios more than 43% at their age bins. Age vs Exited graph points that seniority has a negative correlation with churn.

Thanks to Tableau tooltip, adding balance to the tooltip help you to see which age bin has more balance. Balance is a critical point for the banks, strategically begin reducing churn with the age bin that has a higher balance effect. In this case, 30–40 years bins have a higher balance amount.

Correlation Matrix

There are plenty of ways of predicting churn and calculating the correlation. In this post, I will share a basic sample to see the correlation. Seaborn is a fancy visualization library with many different color palettes. Also, you may get the design with fewer codes.

Another example of correlation visualization;

Basically, the correlation coefficient quantifies two variables’ relationship between -1 and 1. Zero means there is no linear relation, -1 indicates strong negative relation, and 1 indicates a strong positive correlation.

This correlation matrix shows that age and churn have a positive correlation and they have a stronger relationship rather than the other variables. However other variables’ correlation coefficients close to zero which means they have almost no relationship with churn.

Thank you for reading. If you have any questions, you can reach me on Linkedin.

Resources :

https://medium.com/@szabo.bibor/how-to-create-a-seaborn-correlation-heatmap-in-python-834c0686b88e