A machine learning approach was used to analyse and categorise insurance risk events, helping improve consistency, reduce duplication and enhance predictive risk insights.

Issue

This leading insurance-focused operational risk member organisation was looking to enhance the efficiency and effectiveness of its insurance operational risk capabilities through the anonymised and confidential exchange of data about risk events between firms.  

Solution

We helped the client assess whether AI techniques could better categorise risk events. We designed a machine learning model using R as the underlying programming language. Crowe applied language analysis to risk event descriptions and used unsupervised learning to let the algorithm categorise events independently. We also mapped patterns from a labelled dataset to the categories identified through unsupervised learning.

Outcome

Crowe analysis quickly revealed inconsistencies in existing categories, identified opportunities to simplify duplicated and overlapping events, and demonstrated how predictive forecasting supported more effective risk event categorisation in the future.