Deployment of a Biometrical Risk Prediction Model for a Public Insurance

This use case article explains how we support a public insurer in the operationalization of its first Machine Learning (ML) model. From understanding their current application process to designing the most suitable MLOps solution, and integrating it into the running process, we made it possible for our client to improve the speed and the efficiency of its main activities, at scale.

• MLOps
• Data Science
• Machine Learning (ML)
• Production
• Data Engineering
• Business Process
• Insurance

Key Challenges

Our client offers disability insurance (for people who are no longer able to work) and term life insurance. They receive many applications with detailed questionnaires filled out by applicants which require a risk assessment.

We worked with our client on developing and training a Machine Learning model able to semi-automate the process of assessing biometric risk during the application process. Thus, we proved that it was possible to improve the speed and the efficiency of the application process and to increase the processing rate of all applications. But to achieve these results at scale, the model needed to be deployed and integrated into the existing insurance process.

In fact, our client had never moved beyond the POC or MVP stage with a model before. This was actually the first time they wanted to put a model into production. And it’s important to note that deploying a model in an industry as regulated as public insurance is no small task.

Therefore, our client was looking to:

• Get a solution for putting the previously developed model to production;
• Make sure that the model is monitored and audit-proof;
• Be supported in the whole process and get expert advice in MLOps activities.

Our Approach

In order to meet our client’s challenges and therefore bring the model into production, we:

• Took the time to collect information about the current application process applied, the data being used and the ML model that was developed in order to get a clear understanding.
• Designed the relevant solution ensuring a smooth and robust integration of the ML model in the application process.
• Developed, tested and implemented a Webservice and production-ready scripts for this first-time operationalization.
• Implemented security logic to facilitate the control over critical cases via parametrized threshold levels.
• Integrated these into the existing application landscape as well as the developed model and we provided support during the go-live phase of the project.
• Setup a logging and monitoring component and took care of its implementation within the complete solution.

Benefits

• First successful operationalization of a machine learning model for the public insurer.
• Lots of learnings for future MLOps activities and the deployment of new models.
• Great reduction of time-consuming application processing as well as increased customer satisfaction.
• Successful real-time deployment for each life insurance application.
• Audit-proof operationalization of machine learning models.
• Successful integration into the productive process of the application system (dark policing).

Teams involved on this project

A Data Scientist and a Data Engineer collaborated with the insurance expert for 18 months to build this AI-based Application.

Technologies

Share

Thanks to our
Author

Michael A.