BANK FRAUD
Our Human Learning + AI Based Analytics Allowed A Leading Nationwide Bank To Drastically Reduce False Positives, By Detecting 93% Of Fraud And High-Risk Lending. We Can Do The Same for You!
A leading nationwide bank was under pressure to increase its performance on unsecured loans. As with many banks, their credit scoring model wasn't enough to accurately underwrite risks. They were writing off large amounts of debt each quarter, and needed a more effective means of determine risk factors.
BENEFITS
SMARTER FRAUD DETECTION
Our multi-layered approach to detect and prevent fraud, allowed us to identify 9 out of 10 fraudulent transactions in real-time.
REVENUE
Bank saved millions each quarter, by passing on high risk and fraudulent borrowers.
INSIGHT
Bank was able to pinpoint characteristics of bad loans vs. first-party fraudsters clearly
PROCESS
The insights from the project enabled bank to revamp the loan application and approval process.
COST
Multiple start-to-finish time and cost saving benefits. From application process, to detection, investigations, case management, and remediation.
THE CHALLENGE
A leading bank was seeing an increase in the number of borrowers who were applying for unsecured loans, with no intention of paying them back. After millions of dollars in write-offs, they turned to Dextro Analytics to devise a proprietary tool that would help them differentiate between first-party fraud vs. bad debts.
Solution
One of the key challenges of this project was the similar characteristics between first-party fraudsters and genuine customers with bad debts. However, we developed an advanced graph analytics model that allowed our client to accurately detect the minute differences, and fraud in a timely manner.
Our decision and data engineers created a relationship database, which successfully integrated a broad range of internal and external data. We were able to assess risk and fraud with a combination of transaction patterns, borrower behavior, repayment and no-payment patterns,
source of loan applications, demographics, contact information, and relationship among each group of individuals.
Our engineers ran 1000s of tests, on 5000+ parameters to develop the initial data required to move forward. Unlike traditional analytic tools, our innovative engineers visualized the data in graphs that identified patterns across multiple touchpoints, such as:
1. Loan application lead
2. Initial application and account data
3. Crucial events, transaction thresholds, and sudden no-payment
In addition, our decision and data engineers integrated human learnings, and static rules to further build a robust fraud detection model.
RESULTS
1. Built real-time operational tool that helps underwriting detect first-party fraud
2. Saved millions of dollars in fraud/lost debts by detecting more than 90% of potential fraud
“Best-in-class methodology to detect the fraud. Highly innovative”
- Chief Marketing Officer
Our team would love to share the nuts and bolts of the study and our best practices.
Please provide us with your contact information.