Identifying fraudulent checks in real-time
Increased percentage of fraudulent checks identified by system,
hence reduced fraud losses and need for manual review.
Process
Even with lower check-processing times due to electronic payments and automated clearing house (ACH) transactions, banks must still manually verify millions of handwritten checks. Annually, banks risk losing millions as a result of check fraud by counterfeiters. Because a percentage of the funds is made readily available to the depositors, it’s critical to identify counterfeit checks quickly. To reduce the incidence of check fraud, a global bank partnered with us to build a solution based on Artificial Intelligence (AI) machine learning to speed up check verification.
Solution
Solution needed to identify fraudulent checks in real-time, as well as reduce the number of checks requiring manual review. This was achieved using Optical Character Recognition (OCR) and deep learning technology to scan checks, process data and verify signatures. Our model, based on Google TensorFlow™, uses a neural network to parse a historical database of previously scanned checks, including those known to be fraudulent. We trained the neural network to use a set of comparative algorithms to distinguish good checks from anomalous ones. By automatically comparing various factors on scans of deposited checks to those in the database.
Challenges Addressed
Manually intensive check review process
Human error in identifying fraud due to the manual process was a key factor to look at automation.
Signature verification
Standard automation was not effective for identifying signature mismatch hence Computer Vision and machine learning technology was needed.
High volume
Volume was ranging in millions of checks being processed by the bank with low Turnaround Time (TAT) required.
Highly analytical in nature
Fraud detection required high level of analytical skills based on experience for the fraud detection team, this made it difficult to ramp the team size.
System availability
Daily closing cycles reduced system availability to less than 16 hours a day and difficult cut off times. All checks in the system were required to be fraud checked before the closing cycle run and hence sampling techniques were deployed for fraud check.
Outcome
- 50% reduction in fraudulent transactions.
- $5 million annual savings on fraud losses.
- Lower manual check validation operating costs.