TROV, AI-Powered Fraud Detection
AI-Powered Fraud Detection
About the client
Trōv is a global leader in embedded insurance, powering the future of digital insurance distribution and emerging mobility. Its robust insurtech platform empowers financial institutions and insurance incumbents to easily embed insurance products within other digital experiences to increase recurring revenue.
Project info
-
Client
Trōv
-
Industries
Insurance -
Offerings
Solution engineering -
Expertise
Artificial intelligence and data science
Challenge
Insurance companies spend several days to weeks carefully assessing a claim, however, the insurance business is still affected by fraud. The most common issues are property damage, car insurance scams, renters, and pet issues. Carefully selected models may help to successfully detect fraud. Our client aimed at automatic identification of fraud activities and preventing fraud-related losses and damages before claims are paid, and before fraud activities could do any harm to their system. The client also wanted to decrease the time spent for claim consideration and payments to genuine customers.
Solution
Based on Artificial Intelligence and statistical techniques, our team built a robust fraud detection system.
As the first step, before the labeled data was gathered, the Avenga team leveraged an unsupervised clustering approach to identify suspicious groups of users. We gathered all the vital metrics that signal potentially suspicious clients, including ID checks, a PEP/Sanctions/Deceased Warning, Forwarding address link warning, Email/Mobile Risk, credit card scores, etc. In addition, feature engineering was applied which leveraged the domain knowledge in order to manually create new important features that could affect the fraudster’s identification.
At the later development stages, having collected the labeled dataset, we applied a supervised machine learning approach to make fraud-related predictions that helped to boost the accuracy of the fraud detection system. Neural networks were used that can learn suspicious patterns from samples and later detect them in new unseen cases.
Results
Fraud detection analysis helped to minimize fraud, improve underwriting and enhance risk management.
The client benefited from:
- Automatic real-time fraud detection
- Improved accuracy of fraud detection, false positives reduction, and the minimization of human errors
- Costs optimization of the processes and resource usage
- Lowered loss ratio, fraud or break identification before any harm is done
- Accelerated payments on genuine claims
- Improved customer experience
Technology used
-
Python
-
NLP
-
K-means
-
DBSCAN
-
t-SNE
-
Random Forest
-
SVM
-
Neural Networks