Large banks, finance services and insurance firms manage lakhs of customers, who bring in thousands of crores worth of assets with them. It is essential for these businesses to engage customers by providing them with new products by upselling to existing customers or increase customer retention or acquire new customers to continue to generate more revenue. This makes machine learning the best weapon in the retention arsenal and churn prediction the most widely used big data use case.
Corporations spend a lot of money in developing strategies to fight customer from disengaging. Using churn prediction, it will now be possible to detect those customers who are likely to disassociate in the coming month and hence employ all efforts towards those identified, rather than just shooting in the dark.
- Inability to gauge which customers are likely to discontinue.
- Inability to employ individualized customer retention strategies, thereby not fully getting the return on investment on such strategies.
- Not being fully able to leverage ROI on strategies to increase revenue in and around customer engagement.
- Machine learning refers to techniques used wherein intelligence is acquired by referring to examples. In order to predict which customer is going to leave, their historical data is compared with examples of customers who stayed or left.
- Customers display behavior based on their usage of services provided by a business. Hence in order to predict whether the customer will churn or not, we need to gather data points that display this behavior. These could be their basic information, how frequently they communicated with customer service, how they used the service, and any other contextual information related to the customer.
- Using traditional methods like decision tree, random forest, logistic regression it is possible that sometimes random outputs are received since the core of the decisioning is to be done based on the features extracted for custom behavior. classNameical machine learning cannot accommodate the vastness and nuances of customer data.
- Deep learning techniques for classNameification gives better results. We have used artificial neural network to derive our prediction. Advantages to go with deep learnings include getting rid of feature selection process, and let the model derive data points that are of utmost importance in making the prediction.
- In our sample dataset we have identified various data points to be collected like :
- The model was trained on 8000 customer data where in the model was given all these customer data points mentioned above as well as the results whether they churned or not.
- The model was tested on 2000 customer data where in the model was given all these customer data points mentioned above, the model predicted whether the customer would churn or not and it was analyzed against the actual data point for churn. The results are displayed in the confusion matrix below :
- 86% accuracy was achieved, when the model was tested.
- In order to achieve this type of success in solving real world problems data points such as, but not limited to the following, can be used :
- Customer profile
- Bill payments
- Product portfolio
- Usage frequency
- Marketing campaigns tried previously
- Automatic churn prediction for all customers by machine periodically.
- No delays due to manual prediction based on analytical reports.
- Impactful difference in retention before and after churn.
- Taking proactive action beforehand to improve success rates.