Churn analysis

The churn rate is the proportion of clients who do not renew a subscription or a contract. We offer powerful analyses to profile clients and identify those who are more likely to leave.
Below is an example of what we can do to model the churn rate.

1. INITIAL QUESTION

The director of a bank wants to identify, amongst his clients, those whore are most likely to close their account.

2. DATA COLLECTION

Data are collected and anonymised by the bank. Duplicates and incomplete entries were removed from the dataset. Categorical data (Country, Gender, etc.) were encoded as numerical variables for modeling purpose.

3. DATA ANALYSIS

We build a deep learning algorithm, using a network of artificial neurones with 2 hidden layers of 20 neurones each, and a rectifier activation function.
In this simple model, and for the demo, we do not use any k-fold or cross validation, but for a real use the model would be much more sophisticated and optimized.

  

4. RESULTS

The model accuracy is 0.858, meaning that predictions are accurate in more than 85 % of the cases. The results of the model show that the most important factor determining the churn rate is the number of financial products owned by a client.
We deliver a custom application to the director of the bank. This app allows to easily estimate and visualize the churn rate of a client based on his profile. Here is a short video showcasing the app.

Churn bank DataTailors example

5. CONCLUSION

Our model (simple for the demo) is able to detect with an 86 % accuracy weither or not a client is likely to leave the bank. We recommend to give a special attention to the number of products owned by the clients.
An hypothesis is that when a client owns more product, he is more experienced with the banking system and could consider that its current bank is too expensive or not competitive enough.
We deliver an custom application allowing to detect weither a client is likely to leave the bank based on its profile.