Client story

National health insurance - optimising revenue by reducing churn

Situation summary and business challenge

Our client, a major, growing player in the field of health insurance, has one main objective: increasing revenue. There are several ways to do this. Two of them are (1) acquiring more customers and (2) reducing churn. Although keeping customers loyal is becoming a real challenge in today’s health insurance sector, it is still easier and cheaper than acquiring more customers.

"Using the data currently available, how can we reduce churn to most effectively increase our revenue?"


Our proposed solution

We were asked to dig into our client’s customer data to uncover the triggers for churn. The final goal is to incorporate the churn model into their CRM system to ensure that every customer gets assigned a churn score, which will be regularly updated. This score will be used when interacting with the customer (through mail, phone, web, F2F) to personalise communication messages and target marketing campaigns.

As with any analytical project, whether they involve big data or not, we started with a full understanding of the client’s needs, more particularly about which part of the churn they wish to minimise (given that there are different levels of customers) and how they want to proceed.

The next step was to obtain and examine all the known information about the customers and those who left.

A detailed exploratory (and data merging) phase – essential to avoid using unnecessary inputs and overfitting issues – led us to the predictive modelling stage where we tried several statistical approaches. At boobook, we believe in a White Box (versus Black Box) approach. This means that we want to understand the input and output of the models and how the model is built. Otherwise, the model cannot be explained to the people using it later on, plus there would be a risk of overfitting/unreliable models and instability over time.

As with all predictive modelling, we split the available data into a modelling and a validation data set (multiple times) to validate the accuracy and stability of the model.


"Using the data currently available, how can we reduce churn to most effectively increase our revenue?"

Results and benefits to our client

We are now completing the POC phase. The next step is the incorporation of the application as a decision-making tool in the CRM system.


Photo by MGDboston at