ArmorDoc: Challenging the Mortgage Industry’s Status Quo with Intelligent Document Processing (IDP)

Follow ArmorDoc on :

Mike Hartman CEO
What Is IDP And How Is It Different Than Existing Document Solutions?

Existing document solutions rely heavily on OCR (Optical Character Recognition) combined with a rule-based approach which breaks down quickly once the format of a document changes. With IDP, OCR is just one component of the overall solution. The real differentiator with IDP is its reliance on machine learning models. Machine learning relies on being fed high volumes of documents, with high variance, for it to become smarter and more flexible allowing it to continuously improve its accuracy, unlike a rules-drive approach.

A collection of mortgage documents is referred to as a loan package. The biggest problems for key players in the mortgage industry are incomplete loan packages, especially critical documents, and the inaccuracy of the data extracted from these documents. Loan packages change hands from Lenders to the likes of Investors, Servicers, Trustees, Custodians, etc. Each time these loan packages change hands, documents tend to get lost in the mix.

It’s imperative all counterparties have full confidence they are not only receiving the critical documents but are receiving the “final” versions of these documents. In addition, it’s imperative the extraction of loan, borrower and property data from these documents is, not only available, but highly accurate thus adding to the importance of the existence and identification of these documents.

These loan packages can range from 250 to 15,000 pages, or even more based on the age of the Mortgage, resulting in hundreds of millions of pages when loan packages are received in large batches. A wide array of variations and formats exist depending on the year of creation, government entity in which the document was issued and/or company specific generated documents such as title and insurance providers. The reason this is noteworthy is this causes documents to vary in formats resulting in unstructured documents.

As mentioned earlier, prior to ArmorDoc, processing documents was mostly solved by OCR and rules-driven approaches only to be followed up with humans to clean up the poor results. This approach falls apart quickly with unstructured documents thus resulting in inaccurate results. Leaving humans to sift through millions of pages, logging their findings into a dashboard all while using the wizardry of the likes of stare and compare and/or data entry. This approach is laborious, involves human error and painstakingly slow.

The answer to all these document processing issues in the mortgage industry is ArmorDoc.

Being someone who has dealt with these laborious, antiquated processes, Mike Hartman, founder and CEO, launched ArmorDoc in 2021, combining machine learning with serverless cloud solutions to automate the process of document classification and data extraction for forward, as well as, reverse mortgage documents.

“Spending the past 15+ years dealing with these issues without any viable solutions along with feedback from prospective clients were enough for me to launch ArmorDoc. Our clients are the ones with a vendor having sold them vaporware or those clients who have attempted to build an inhouse solution. I’ve made a conscious decision for ArmorDoc to remain laser-focused on solving document problems and let others focus on what they do best,” says Hartman.

ArmorDoc is a leading data science company leveraging the power of machine learning to solve document classification, data extraction, and stamp/signature recognition problems. The company’s superior, flexible and scalable AI converts clients’ mortgage documents ranging from hundreds to millions of pages into usable data at scale.
To do this, ArmorDoc combines machine learning with serverless cloud solutions to classify, organize, and extract data from large electronic mortgage documents with speed and accuracy, making it ideal for processing at scale. With this unique approach, ArmorDoc’s findings get fedback into the machine learning process to improve the accuracy of its automated models. The solution’s ability to provide 99 percent accuracy in document classification and 97.5 percent accuracy in data extraction eliminate the need for manually manipulating the results prior to delivering them to clients. As a result, the company can expedite results back to its clients quickly, resulting in an 80 percent reduction in processing timelines, 90 percent reduction in postclosing trailing documents, and a 50 percent increase in cost efficiency all while achieving 95+% accuracy.

Designed to address the modern-day requirements of the mortgage industry, ArmorDoc runs every page of a PDF document through its pipeline which is comprised of a comprehensive seven-step process. This is ideal, as mortgage documents can be uploaded in large volumes with some having many as 15,000 pages. It also comes with an image rotation model that rotates images to their correct orientation, given the possibilities for sideways and upsidedown pages in mortgage documents.

Our proprietary AI solution is a valueadded alternative to the decades-old status quo in document processing—OCR— that results in high inaccuracy and large postprocessing QA effort by a group of humans, mostly offshore

“Our proprietary AI solution is a value-added alternative to the decades-old status quo in document processing—OCR— which results in high inaccuracy and large postprocessing QA effort by a group of humans, mostly offshore,” states Hartman.

Illustrating ArmorDoc’s value proposition is its collaboration with a client which was purchasing a portfolio (~140,000) mortgages from a counterparty. Without ArmorDoc’s offerings, the client had no insight into the quality of documents prior to the trade closing. The client would have spent months alleviating post-closing complexities.

ArmorDoc alleviated most of the post-closing pain which currently exists today, allowing both parties to focus on future opportunities, instead of being bogged down by previously closed trades.

In another instance, ArmorDoc engaged with an origination company, integrating its solutions into their workflow to automate their pre- and post-origination QA process. The company was able to alleviate the need for the client to ramp up and scale resources, allowing the workforce to leverage their skills on making important underwriting and risk management decisions, instead of clicking through PDF files.

These success stories are a testament to ArmorDoc’s core focus on document classification, data extraction, and stamp/ signature recognition, while its competitors consider this an add-on or additional service. Its ability and willingness to take part in champion challenges and perform live demos with never seen before PDF files provided by clients at the time of the demo is also a key differentiator of ArmorDoc.

With all these capabilities, ArmorDoc is building a unique niche in the mortgage industry for large bulk transactions or in creating workflows via API integrations with existing service providers catering to the same client bas.


New York, New York

Mike Hartman CEO

ArmorDoc is a leading a Data Science company that leverages a myriad of machine learning techniques to solve document classification, data extraction, and stamp/signature recognition problems. The company’s superior, flexible and scalable AI identifies missing documents and converts millions of pages into usable data at scale with speed and accuracy