Case Study
Historic (Liber-based) Real Property Data Capture
Challenge
A large financial institution needed historic real property information for a home affordability analysis. Originally contained in liber volumes, the pages had since been digitally scanned to .pdf images. The low-resolution images were of poor quality, including wrinkled pages, skewed columns, mis-aligned rows, etc. Given the large volume of data (>50MM characters), manual data entry was not feasible.
Industry
Financial Services / Banking
Data Type
Semi-Structured Text
Project Duration
5 Months
Ongoing?
No
Solution
Our solution included multiple phases: initial staging and preparation of the data, automated scanning and capture, quality assurance checks, and output formatting. Data Associates initially parsed each scanned page image to enable enlarging the image to reduce the density of the data. An AI-driven scanning process captured 50-70% of the contents of each page, augmented by human-in-the-loop review and correction to increase the capture rate to 99%. A two-step quality review verified the final captured information against the original image to bring the capture accuracy to the required 99.5% level. Finally, the captured information was formatted to match the input requirements of the analytic platform.
Outcome
The multi-stage intake process, combined with human-augmented automated data capture allowed for a solution that complied with all requirements – including a faster throughput speed than was anticipated
Download Case Study
