Axis Technical Group (“Axis”) and an industry-leading financial services firm (“Client”) completed Phase 1 of an Automated Loan Classification, Separation, and Data Extraction Project. The Client needed a way to streamline their due diligence and audit process by organizing their loan packages — ranging from 250 to 5,000 pages – into a consistent document format. This included classifying, separating, and reordering the loan documents and extracting the necessary data to perform the audits.
Axis’ goal was to allow their financial services Client to quickly locate and pull data from the loan documents with ease and in a repeatable fashion, resulting in roughly 10-15% time savings and $2.32 to $4.37 cost savings per loan, depending on the audit type.
The Client processes thousands of loan files each month – ranging from 250 up to 9,000+ pages in length – which are received as a single PDF.
Their previous audit process required $27/hour underwriters to re-underwrite the loan for Credit, Compliance, and Forensic reviews. This means the auditor must scroll thru the 250 to 9,000 pages of the loans numerous times looking for specific information on a particular document.
Then they would have to manually data enter 150 to 300 fields (depending on the review type) into the LOB system to perform the audit. Since the documents in the loan were in no consistent order, this was an extremely time-consuming process and not an advantageous use of the underwriter’s time and expertise.
This typically would take 2.5 to 3 hours per loan and could not be done by lower-cost data entry clerks because the underwriter is required as an SME (Subject Matter Expert) to identify the correct information from the document. Alas, one of the nuances of the mortgage loan servicing and compliance process.
Axis applied its AI-based data extraction solution to automatically classify, staple (separate), reorder, and bookmark the loan by the document type (i.e. Note, Mortgage, Deed, Income Documents, etc.). This process had to be completed with a consistent, easily digestible format that allowed the underwriters to locate the information faster while increasing the speed of data entry.
The images are processed through an optical character recognition (“OCR”) process, after which the OCR’d data and images are classified and analyzed using Axis AI-based managed service utilizing patent-pending Natural Language Processing and Machine Learning technology. The required documents were then classified, stapled, reordered, and bookmarked in a specific order to streamline the data entry into the LOB system.
Documents submitted by our financial services Client were processed and data indexes were delivered back to the Client in less than 48 hours. The project recently reached completion on schedule and on budget.
Phase 2 for this financial services
The next steps will be to automatically extract the 150 to 300 data elements and programmatically feed the LOB system. The required data elements will be extracted, validated, and/or corrected, and the values then returned to the Client for seamless ingestion into their LOB system.
This will not only eliminate most of the manual data entry but will allow a lower-cost data entry clerk to validate the data. This frees up the underwriters and allows them to focus on their primary purpose, which is to underwrite the loan.
This combined savings in time and labor cost will result in an estimated 54% increase in throughput, which equates to $8.11 to $15.73 total savings per loan.
The Managing Director for our global partner had this to say about the project:
“The Axis Team is outstanding, to say the least, and we truly appreciate everything they have done to enable us to streamline our process! We are mostly pleased with the 15% savings from Phase 1 alone, and want to ensure that all future eligible loans are sent through the Axis AI engine.”
Axis’ CEO, had this to add:
“We are excited to see our NLP technology utilized on a broad range of disciplines and document types, from structured to semi-structured and the historically challenging unstructured document. Unstructured content is what the Axis AI engine was designed for, but our clients are finding it also incredibly efficient on structured content, like tax forms.
Axis’ utilization of AI is perfect for a business process that encompasses a mix of document types and layouts, where clients don’t have to mess around with bar codes, separator sheets or manual sorting to classify and prepare their content. The labor savings are incredible in both time and money.”
Beware of Imitators
Historically, from what we’ve heard from clients, many companies tout their ability to do execute a project like this. But as production crashes, it becomes apparent to the client that they should’ve gone with someone who specializes in classification and extraction for various document types. Though most firms find it difficult to capture data from unstructured documents, Axis has made it easy.