AI-based Fraud Detection

As part of the AI-based Fraud Detection project, b-nova supported Helvetia Insurance Switzerland in the technical and domain-specific implementation of automated fraud detection. The focus was on analyzing relevant data sources, developing scoring models and detection logic based on machine learning, and preparing a robust integration into existing claims processes and business systems. Additionally, we accompanied data preparation, systematic testing, and the iterative optimization of detection quality, taking into account false-positive rates and domain-specific thresholds.

Biggest challenge

Translating domain-specific fraud patterns into robust technical detection logic with heterogeneous data quality while minimizing false positives

What we did

Analysis, design, and implementation of an ML-powered scoring and detection solution for automated fraud detection including data preparation, system integration, and operational readiness

Main tools we used

Python, Java, Machine Learning, REST, Data Analysis, OpenShift, Splunk

Tasks

Analysis of business requirements and close coordination with stakeholders to define detection goals and fraud patterns

Identification and evaluation of relevant data sources from claims, contract, and customer systems

Data preparation, cleansing, and feature engineering to support detection logic

Design and implementation of scoring models and rule-based detection mechanisms

Development and training of machine learning models for anomaly and pattern detection

Setup of a reproducible ML pipeline for model training, evaluation, and deployment

Integration of the detection solution into existing claims processes, interfaces, and business systems

Connection to third-party systems and business applications via REST APIs for real-time data queries

Systematic testing and validation of detection quality including false-positive analysis

Iterative optimization of thresholds and scoring parameters based on domain feedback

Setup of monitoring and alerting to track detection performance in production

Containerization and deployment of the solution on OpenShift for scalable provisioning

Setup and maintenance of CI/CD pipelines with GitHub Actions for automated builds and deployments

Comprehensive documentation of architecture, detection logic, and operational processes

Knowledge transfer and coordination with business and development teams for sustainable adoption

Technologies

Python Java REST Jupyter / Data Analysis Machine Learning OpenShift GitHub Actions Splunk SQL / Data Preparation API Integration