Finally, AI and machine learning made for mortgage!

Artificial intelligence (AI) and machine learning (ML) are everywhere these days.

From generative AI creating videos, music, written text and presentations, to large language models understanding and processing natural language, there’s a lot of cool stuff out there. As mortgage professionals, it’s tempting to apply these tools to mortgage-specific tasks. However, using generalized AI tools for mortgage purposes is more complex than it may seem.

This doesn’t mean there aren’t great AI/ML capabilities for mortgage lenders. In fact, Dark Matter Technologies has been developing proprietary AI/ML models to streamline mortgage processes since 2018.

AIVA rocks stare-and-compare

Built by a team of dozens of data scientists, our AIVA® Virtual Assistants (pronounced “Āva”) have been many years in the making. Notably, this suite of AI/ML solutions is what’s behind our advanced automated document classification and data extraction capabilities.

As you may already know, there’s nothing basic about sorting over a thousand different document types used in mortgages and extracting meaningful data from those documents. Particularly when looking at less-structured documents, the task becomes downright difficult. It’s no wonder many lenders still do a manual stare-and-compare QC process or manually calculate income off paystubs, W2s and tax returns.

Early solutions using traditional location-dependent OCR tried to make a dent in the problem. Still, the lack of structure in many of the documents used in the mortgage process and the inevitable rate of change of documents had these tools hovering at an uncomfortable 55% to 65% accuracy rate. It’s tough to rely on automation that is only 55% correct! Lenders ended up spending a lot of time checking their automated OCR for accuracy and correcting issues. What’s worse is that all that checking was a point-in-time solution. Correcting it once didn’t train the tool at all — it only trained the human doing the review.

AI/ML technologies are a natural fit for solving this problem. Training a model to “read” documents, identify what they are and then extract key data is a “no-brainer” in more ways than one. The ability to have manual corrections recognized by the model and used in future document reviews seems like a quick path to full automation.

AIVA Docs Shines

AIVA® Docs is a mature, mortgage-specific AI/ML solution for document classification and extraction that is now making a difference for more than 45 lenders. AIVA Docs recognizes more than 1,100 mortgage document types and extracts more than 1,200 key mortgage data elements from documents.

AIVA Docs accepts batch image files and sorts them into individual mortgage document types at a very granular level—think tax schedules vs. tax returns. Each document returned has an XML file accompanying it with metadata about that document, identifying the document type and any document data that was extracted from that document. These data files provide a document-level, structured data set that facilitates document reviews using deterministic rules engines, like AIVA® Rules.

AIVA Docs is fully integrated with the Empower® loan origination system (LOS) but also is LOS-agnostic.

Inviting AIVA Rules to the party

AIVA Rules is an extremely powerful rules engine that, when coupled with AIVA Docs, drives greater efficiency. Used with AIVA Docs, AIVA Rules can identify discrepancies between documents and data already in the loan file, identify what documents are missing from a loan file and determine which documents are missing signatures. AIVA Rules can even review a bank statement and identify interesting deposits or look at the data from W2s, pay stubs and tax return schedules to calculate income.

Suddenly your human resources aren’t bogged down in the more tedious acts of running calculations manually or keying data into calculators. Humans are not sorting documents and eyeballing data from documents and data from the LOS to identify issues.

The issues are already identified for them, allowing them to concentrate on the exceptions, clear those and get more loans back onto the happy path of funding or delivery to investors. Resources so empowered can manage more loans each day, bringing a much more scalable workforce to bear against this cyclical industry.

Adding AIVA Assist to the equation

Recognizing the power of AIVA Docs combined with AIVA Rules, and what a strong workforce assistant they make, Dark Matter Technologies built AIVA® Assist capabilities to provide a framework for these tools to operate.

AIVA Assist processing provides a user experience and workflow where income calculations and bank statement reviews can be seen by those making decisions on loans. AIVA Assist QC provides a user experience and workflow where the review of the collateral package is quickly visible and exceptions can be attacked quickly, allowing lenders to get loans delivered faster.

We’re charting the future of mortgage with AI

I think what Dark Matter Technologies has built over the past five-plus years is pretty impressive. But don’t just take my word for it. We’d be happy to show you how it all works, and then work with you to calculate the return you could expect from an investment in AI/ML, focused on mortgage use cases. We’d be happy to run a proof of concept for you as well, using documents you provide to pass through AIVA Docs so you can see what it can do in a real-world example.

Now is the time. AIVA Virtual Assistants are the answer.

Contact us today for a demonstration.