Introduction
During recent mortgage tech events MBA Annual in Philadelphia , I came across a repeated pattern of questions - how ChatGPT and the whole large language models impact the business of loan document processing for the mortgage industry.
While several document automation solutions already exist in the ecosystem - each utilizing different AI models to a varying degree, there are still ambiguities on these solutions size up to one another. We'll try to resolve some of these ambiguities through a series of side by side comparisons. Eventually, we'll be able to piece together a bigger picture of how AI based document automation could impact loan processing business.
Setting Expectations
To effectively support loan processing, these tools must possess the capability to perform critical tasks. These tasks include -
- Identifying and splitting documents
- Extracting data from structured and unstructured documents
- Simplifying output review for better decision-making.
Software Overview:
AWS Analyze Lending API
Analyze Lending API is a robust document processing API specifically designed for mortgage documents. With Analyze Lending, you can effortlessly extract, classify, and validate critical information from mortgage-related documents. The API intelligently splits loan documents into pages and accurately classifies them based on document type. These document pages are seamlessly processed by Amazon Textract, ensuring precise data extraction and comprehensive analysis for mortgage-related tasks. Experience the efficiency and accuracy of Analyze Lending to streamline your mortgage document processing needs.
Document Types Recognized by Analyze Lending
The table below presents an exhaustive list of document types recognized by Analyze Lending, along with an indication of whether each document includes a signature field.
Vaultedge Document AI :
Vaultedge is a comprehensive end-to-end loan processing solution equipped with advanced document management features. It leverages the power of artificial intelligence and machine learning to extract, classify, and validate data from various types of mortgage documents. With its powerful OCR capabilities and custom-built models, Vaultedge accurately extracts data from structured and unstructured documents. It is purpose-built for the lending industry, utilizing large language models trained on millions of mortgage documents. Vaultedge not only accurately classifies documents but also extracts valuable information, making it an invaluable tool for streamlining lending processes.
Document Types Recognized by Vaultedge document AI
Vaultedge Document AI can accurately recognise and categorize over 500 mortgage document types and extract more than 2000 fields of information.
Side by Side Comparison
Performance Test:
We will conduct a comparative analysis of AWS Analyze Lending API and Vaultedge Document AI for processing lending documents. Our objective is to examine the capabilities of these solutions side-by-side. We will utilize standard documents such as a note, deed of trust, and loan estimate, and evaluate the outcomes obtained from each option.
AWS Analyze Lending API -
Let's dive right into it. Within the AWS console, we have the option to upload a file and examine the output.
First, let me show you the documents that we'll be processing: a note, a loan estimate, and a deed of trust. These documents are fairly standard and familiar to your team members.
Now, let's see if the software can understand these documents as well as we do. Uploading the note to the AWS Analyze Lending API,
we aim for the software to identify the document and extract the relevant information. We're keeping it simple by uploading one document at a time.
After analyzing the three-page note, the classification output from the AWS Analyze Lending API is as follows:
The software indicates that page one contains a payoff statement, while pages two and three are unclassified. This document is meant to be a simple and straightforward note, and the classification results are completely inaccurate.
Now, let's examine the data extraction results. The software successfully extracted certain information, such as the principal balance and loan amount.
However, it completely missed extracting the interest rate, which is clearly visible. We would typically expect the principle and interest to be extracted according to the note, but that was overlooked.
Furthermore, there are some irrelevant fields displayed by the software due to incorrect classification.
In examining other documents, the recurring pattern becomes evident.
In this instance, a deed of trust has been uploaded.
Upon review, the software incorrectly identified page one as a payslip, while pages two to five remained unclassified.
On page six, it correctly identified it as a 103, but once again, pages seven to nine could not be identified. Finally, on page ten, it was again identified as a 103.
The classification results are significantly inaccurate, which is also reflected in the data extraction. This was observed using the AWS Analyze Lending API.
Vaultedge Document AI
In Vaultedge, we have the convenience of uploading all documents at once, eliminating the need to upload them one by one as in AWS Lending API.
Now, let's examine the indexing output. During this process, the software accurately identified the first three pages as a loan estimate, followed by the next 16 pages categorized as a mortgage deed of trust. Finally, the last three pages were identified as a note.
Moving on to data extraction, Vaultedge impressively extracted a significant amount of information from the deed of trust, including the property address, MERS ID, borrower, co-borrower, and more.
The software also provides a confidence rating, allowing for efficient review of low confidence items. By leveraging Vaultedge, time is saved by focusing only on the two pieces of data flagged as low confidence.
Similarly, for notes, Vaultedge extracts a wealth of information, providing the option to review low confidence items if desired. This stands in contrast to AWS Lending API, which failed to extract the interest rate.
Vaultedge's accuracy and comprehensive data extraction capabilities make it a valuable tool for mortgage document processing.
Conclusion
Based on my comparison, it appears that the AWS lending API is primarily designed for structured documents such as W2 or tax returns. However, in the mortgage industry, the majority of documents are unstructured or semi-structured, including notes, deeds of trust, and homeowners insurance. However, these APIs may not seem suitable for handling such documents effectively.
On a final note, if you are in the mortgage industry and exploring high accuracy document automation solutions for semi-structured & unstructured data, then let's chat to see what could be better alternatives to your current approach.
Schedule a call with us today and let's find the best solution for your mortgage document needs!