Customer churn prediction in banking
9 likes10,932 views
The document outlines the establishment of a customer churn prediction model for an Icelandic bank, defining churn as customer inactivity or account closure. It identifies key variables such as demographics and behavioral data that should be considered for accurate predictions, recommending either decision trees or logistic regression as suitable modeling techniques. The conclusion emphasizes the importance of targeting customers who have not been active for three months and utilizing these methods to potentially minimize churn rates.
Customer churn prediction in banking
- 1. Customer churn prediction model in banking Recommendation on how to set up a customer churn model for an Icelandic bank
- 2. Customer churn in banking • Churn is defined as movement of customer from one company to another. The reasons can for example be: • Availability of latest technology • Customer-friendly bank staff • Low interest rates • Location • Services offered • Churn rate usually lies in the range from 10% up to 30%. 2
- 3. Customer churn model • Prediction models are used to identify customers who are likely to churn • The model uses historical data on former churners and tries to find some similarity with existing customers • If some similarity is found those customers are classified as potential churners. 3
- 4. Why is this important for the bank? • The cost of attracting new customers can be five to six times more than holding on to an existing customers • Long term customers become less costly to serve, they generate higher profits, and they may also provide new referrals • Losing a customer usually leads to loss in profit for the bank. 4
- 5. What needs to be considered when setting up a churn model • How to define a churner in the bank • Churn prediction variables to use in the model • Methods/techniques used to build the model 5
- 6. How to define a churner in a bank • Customer who closes his account or has decreasing number of transactions over a specific period in time • Focus on customers who have three or less products with the bank 6
- 7. How to define a churner in that Icelandic bank • Active customer is a customer with two or more active products • A churner can be defined as a customer who has not been active over a specific time • One of the active products is an account 7
- 8. Churn prediction variables • Four types of variables that are mostly used: • Customer demographics variables • Perceptions variables • Behavioural variables • Macro environment variables 8
- 9. Customer demographics variables • Most used variables in churn prediction • Age • Job type • Gender • Family size • Geographical data 9
- 10. Perceptions variables • Perceptions variables try to measure how the customer appreciates the services or the products. • Include dimensions such as: • Quality of services • Satisfaction with the company • Locational convenience • Pricing 10
- 11. Behavioral variables • Behavioral variables look at the previous behavior of the customer • How often he uses the services or products • Which services or products he uses • Most popular behavioral variables are number of purchases and money spent 11
- 12. Macro environment variables • Macro environment variables focus on identifying changes in the world that could affect the customer. 12
- 13. Churn prediction variables – what to use in that Icelandic bank • Demographic variables and behavioral variables • Available in the bank database 13
- 14. Methods/techniques used to build a model • Data mining classification techniques • Neural networks • Logistic regression models • Decision trees • Survival analysis 14
- 15. Methods/techniques used to build a model • Neural network • Performs very well • Does not express the uncovered patterns in the underlying data in an easily understandable way • Often being thought of as a black box • Tend to be relatively slow • Can be a tedious process to make the model learn 15
- 16. Methods/techniques used to build a model • Logistic regression • Can give very strong insight into which variables are likely to predict the event outcome 16
- 17. Methods/techniques used to build a model • Decision trees • Easy to use • Shows which fields are the most important • Can be vulnerable to noise in the data • Leaves in the decision tree could have similar class probabilities 17
- 18. Methods/techniques used to build a model – what to use for that Icelandic bank • Use either logistic regression or a decision tree • Both identify the most important variables • Neural network is usually thought of as a black box 18
- 19. Factors that could influence a churn in that Icelandic bank • The dataset: • All active individuals how are customers of the bank • Customer who has not been active for the last three months are considered as a churner • Only 18 years and older • Customers who were deceased were removed from the dataset • The data used was based on the period off 01.Jan 2015 to 30.Sept 2017 • Based on this about 9% of the banks customers where considered churners 19
- 20. Hypothesis one – Younger people are more likely to churn This hypothesis is true 20
- 21. Hypothesis two – Customers who belong to a branch that has been modernized are more likely to churn • Modernized branch – one cashier and more self-service machines • This hypothesis is true 21
- 22. Hypothesis three – Customers who live in the capital are more likely to churn • The option of transferring to a different bank is somewhat limited in the countryside • This hypothesis is not true 22
- 23. Conclusion – my recommendation • A churner should be defined as a customer who has not been active for the last three months • Behavioral and demographic variables should be used as an input for the model • Decision tree or logistic regression should be used as a method/technique 23
- 24. Questions? 24