By continuing to use this site, you agree to our use of cookies. Read our Cookie Policy and our Privacy Policy for reference.
Fraud Detection - Policy Similarity Checking
Triglav, an insurance company, sought to automate the identification of misrepresented policy renewals using Machine Learning (ML). The solution, developed on IBM Cloud and Watson, involves feature creation, an ML model for policy pair comparisons, and a rule-based system for accuracy. Migrated to a new environment and updated, the automated pipeline accelerates the identification of potentially misrepresented policies, improving efficiency and facilitating human validation where needed.
Services
Data Science
Project Length
1 Months
Client
Triglav
Our Planning Process
Insurance agents at the insurance company Triglav get incentives when they sign a new policy as opposed to just renewing an old one. This can cause some renewal policies to be made like they are new to collect the benefits from a new policy. The goal of the insurance company is to detect such policies. Up until now they checked this manually at random, and this process took a lot of time. The idea was to automate this process with Machine Learning and potentially similar policies (I.e., it is a renewal instead of a genuine new policy) to be forwarded to the company.
What we did for Triglav
The first part of the solution was to create features from the policies on which two of them can be compared. The second part was to build a Machine Learning model to predict the renewals. This was performed in policy pairs, where you have one policy which is an old one, and a new one which we are checking if it is in fact a renewal for the first one, just marked as a new policy. The solution was developed on IBM cloud and the IBM Watson platform. To improve the ML model, a rule-based system was also introduced. As final steps, the solution needed to be migrated to a new environment in IBM Watson and update the code to a newer python version, as well as putting the solution into production and automating it.
The first part of the solution was to create features from the policies on which two of them can be compared. The second part was to build a Machine Learning model to predict the renewals. This was performed in policy pairs, where you have one policy which is an old one, and a new one which we are checking if it is in fact a renewal for the first one, just marked as a new policy. The solution was developed on IBM cloud and the IBM Watson platform. To improve the ML model, a rule-based system was also introduced. As final steps, the solution needed to be migrated to a new environment in IBM Watson and update the code to a newer python version, as well as putting the solution into production and automating it.
Final Results
The deliverable of the project was the automated pipeline for checking new policies if they are genuinely new. The solution was deployed on IBM Cloud and IBM Watson and can greatly speed up a manual process of randomly checking if a policy is new or just a renewal. It can check many more pairs of policies and compare them using the generated features extracted from them and then forward the most likely candidates to a human.