Modernizing US Treasury Savings Bonds Records
ABOUT
The U.S. Department of the Treasury (Treasury) assists owners in locating their cash matured, unredeemed savings bonds and confirms there are open bonds associated with their social security numbers or names. With approximately 60+ million in unredeemed savings bonds , the Treasury’s Bureau of the Fiscal Service was tasked with identifying and promoting emerging technologies with the potential to modernize the state of savings bond redemptions and records.
CHALLENGE
Savings bonds are currently stored as scanned images, and extracting the information needed to support locating the owners of the bonds is neither cost nor time effective when done manually. The bond images vary in quality, with some being very old. Additionally, the scanning process often introduces issues such as image noise, image rotation, and multiple bonds or other documents in the same image, such as receipts and reel information, introducing a requirement for classifying different image content.
The extraction of key attributes must meet standards for accuracy. Not every attribute on a bond can be read in an automated fashion at scale but a combination of attributes must be present to determine how to find the bond owner. Prior efforts at machine learning were deemed ineffective so an alternate approach for information acquisition was required.
SOLUTION
Titan Technologies’ Traversable™ Extract solution automates the extraction of 15 key attributes, auto-classifies and sorts bonds by type and year, mitigates issues with image quality, solves the issue of rotated images and images with more than one bond, and supports multiple analysis methods for reviewing bonds by type and year. Validation and benchmarking tools ensure extracted attribute accuracy.
RESULTS
With the implementation of Traversable™ Extract, the Treasury was able to successfully process 60 million images of varying quality in a 7-month window. Traversable ™ Extract exceeded the required attribute accuracy and resulted in decreased labor costs.
