The most important parameter to build a successful Insurance business is the “Ability to Attract, Retain and Grow” the most profitable customers. While a lot of efforts and resources are spent to attract new customers, the attention on Retention and Growth of existing customers is not that high.
Customer retention is essentially the ability to have reduced customer Churn. Customer may churn for various reasons
- Not happy with the service provided
- Competitor gave a better price for a new policy
- The current insurance policy does not fit his or her needs
- Negative feedback on social media, press etc.
- Complaints were not addressed in timely manner
- Company simply fails to communicate and engage with the customer
In most cases, the companies are not fully aware of why customers are leaving and the reasons are anecdotal experiences of sales people. A data and analytics driven approach can help understand the reasons of customer churn and can help predict and prevent it. The approach becomes far more critical if the most valued customers are leaving the most.
Predicting Customer Churn
In a recent customer churn model for a life insurance company, we found that there are over 20 parameters that can impact churn although the weightage of each parameter is different. This demonstrates the complexity associated with a customer churn.
A predictive modelling driven strategy is the best method to reduce churn. Data science and statistical modelling can help predict the customers with high probability of churn. The step-by-step approach to adopt this strategy is as below:
Churn Prediction Process
a)Define the business objectives
It is critical to identify which areas of business are having maximum churn and there is a need to reduce it. A quantifiable goal has to be established to build an ROI. A reduction of 15-20% churn is a reasonable goal in the first year of this strategy.
b)Identify Parameters affecting churn
This is a very critical exercise to ensure a good model is built. A detailed data analysis is to be done to estimate the data distribution and list all the parameters that can potentially impact churn. A separate business discussion should be done to capture the additional parameters that they may share based on the experience.
c) Predictive Model Building
Having a good churn model is important to have a high accuracy. It is advisable to test at least 4 different algorithms and choose the one with best results. Once the model is built and put in production, it should be tested and refreshed regularly- at least once a month during the initial phases and quarterly thereon.
d)Score customers likely to churn
After the model is built, True Positive Rate (TPR) and False Positive Rate (FPR) are calculated to draw an ROC curve. It is a judicious mix of defining accuracy and covering larger population in defining the scoring parameters. Once a “High Churn Risk” customers are identified, appropriate campaigns should be run along with the field force to reduce churn. The business processes concerning all customer touch points are to be re-evaluated for corrective and preventive measures.
Acquiring new customers is 6 times costlier than retaining the existing ones. An analytics driven effort can reduce churn by about 15-20% in the first year itself.
For further information, send a reply to the email or write to the author at dinesh@kloutix.com
