Modernizing Solutions Together

In the health insurance industry, it can be extremely challenging to find useful indicators of unhappy customers outside of direct feedback from customer service calls or surveys. Yet, keeping customer churn as low as possible is extremely important, since the cost of acquiring new customers is steeper than retaining existing ones. Any improvement in customer churn has a big impact on revenue.

Gaining Greater Visibility and Leveraging the Power of Data

Identify members with a high probability of changing providers prior to the open enrollment period based on a limited dataset of claims and demographic data.

Our Challenges
and Learnings

Obtain optimal classification results when working with imbalanced datasets resulting from a much lower number of disaffiliations compared to existing customers. Employ business acumen to balance a dataset by engineering new features without changing the real-life outcomes described by the original dataset.

The Wovenware Approach

We used Innovation Sprints to examine the quality of the data and determine the feasibility of creating a churn prediction model.

Technologies we take advantage of

Building it Right

Initially we did not find any meaningful features to train a deep learning model given that all demographic and claims data followed the same distribution, so here is what we did:

  • Preprocessed data to consolidate claims and performed an exploratory data analysis.
  • Engineered new features to balance the dataset based on knowledge about the client’s data.
  • Employed a deep learning model consisting of three fully connected layers, a single neuron at the output, and the sigmoid activation function.
  • Optimized a binary cross-entropy loss using a sigmoid output.
  • Used a holdout set to validate the model and avoid statistical bias.

❝We succeeded in creating a machine
learning algorithm with a 93% accuracy.❞

Let’s create useful technology together that empowers humans and delivers impact.