Case Summary
The client, a Fortune 500 company, is an American retirement, investment, and insurance company. The client needed a single unified model to provide to uncover more fraud patterns. SLK worked with the client to optimize the existing fraud-detection model spread across various channels, and co-innovated to convert it to a single, centralized model. This led to $11 million in yearly savings, with a 15% improvement in model performance and 30% savings in infrastructure costs.
The Challenge
The insurance client employed individual fraud-detection models across various channels, such as web, mobile, and others. The existing siloed models were prone to runtime performance issues and were not efficient in identifying the complete patterns of fraud possibilities. Managing and maintaining these individual models added to extraneous costs. Therefore, the client was looking for a strong data partner who could unify these channel-specific legacy models, optimize the infrastructure, and improve the model performance and accuracy
The Solution
The SLK team built a new data-science model to identify fraud with parameters that covered multiple channels. The new model considered data variables such as Variety, Velocity, Veracity, and Volume, uncovering all possible patterns for modeling. This model:
- Enabled the pipelining of data from various channels
- Identified the right parameters and ensemble models to enable the seamless optimization of fraud detection, ensuring its accuracy and performance
- Pre-processed data for detection, validation, and error correction, and filled in the missing data or rectified the incorrect data, as necessary
- Eliminated false alarms, estimated risks, and predicted the future of current transactions or users
Business Impact
$ 11M
Savings per year
500%
Reduction in runtime
15%
Improvement in model performance
