Pronounced like the name “Ava,” Dark Matter’s Aiva® virtual assistants are specifically designed and bundled for mortgage origination, helping lenders harness the power of AI, business rules engines and machine learning (ML) to automate manual tasks, reduce cycle times and enhance efficiency in loan processing.
From day one, we knew our AI tools would need to work “as one” with our Empower® LOS platform to turn documents into data that could be support origination processes within the LOS. Our Empower Portals – Borrower Portal, Loan Officer Portal, Broker Portal and Seller Portal – have been designed to send documents to Aiva, so Aiva automatically classifies them and extracts data from them. This capability has grown steadily to where we’re now classifying some 1,100 document types and 1,500 data elements with extremely high accuracy.
So even though lenders using other commercial or proprietary LOS systems may use Aiva, we’ve designed it to work organically with Empower – and soon that will be the case with our NOVA LOS, as well.
A Match Made in the Lab
From the Empower side, many of the LOS’s native capabilities add value on top of the value Aiva provides. Let’s say a correspondent lender uploads an 800-page closed loan file into the Seller Portal, and wants to see how the package was divided into individual document types.
Empower users have visibility into that document package of 800 pages that ended up being distributed into these different document types. They can see how each page from that package was classified. That becomes handy when the seller is saying “hey, that document you’re asking for was in the package I uploaded,” to which our client can then say, “well, actually no, this is how your package broke out,” and provide the details.
Aiva also displays extraction data alongside Empower data, and that data can be used for data augmentation. So in the correspondent lender use case, and the same is true for a wholesale lender, the 3.4 file that they get in is pretty robust, but there’s still data that it doesn’t have, that users can get from documents.
We visualize that to the lender in kind of a side-by-side view so they can see the data that Empower has, and the data that Aiva has extracted from the documents. Then the user can decide if they want to accept the data from Aiva or edit something – all through one user experience.
Intelligently Leveraging Data
While it’s significant that Empower can leverage extraction data, it may be equally important to view the data returned by Aiva at an individual document level and run rules to ensure completeness and accuracy. For example, quality rules from Aiva Rules will broadly compare the data from the documents to the data in the LOS, and might reveal that the note amount doesn’t agree with the loan amount. Next the user would get information back from the Aiva Rules engine, which is the deterministic engine component that identifies discrepancies between documents and all of the data, showing them as conditions in the Empower LOS.
Even more powerful is when one of those discrepancies is automatically cleared by the process. For example, maybe the security instrument is missing and a condition has been created by the system. When the seller uploads it, Aiva knows it’s a security instrument, and clears that condition automatically. The same thing happens for data discrepancies. If a new note is uploaded with the correct loan amount, that condition would be cleared.
In addition to looking for discrepancies, data from Aiva processed through Aiva Rules can help prepare loans for the underwriter. Running income calculations off of income documents, like W-2s, pay stubs, leave and earning statements, tax returns, and a variety of different self-employment documentation helps loan officers, processors and underwriters be more efficient. Aiva can calculate the income and present findings through our Aiva Assist user experience to show, for example, two years tax returns side-by-side, the results of those calculations, and how they were performed, as well as calling attention to interesting findings, like the self-employed income is declining, or there is a significant difference between calculated income and stated income.
That loan file preparation extends to asset statements as well. Aiva automates asset reviews, going line-by-line through the bank statement to identify any potential crypto currency transactions, large deposits or recurring payments, for example, and can even compare that payment data to data in the credit report to identify potential unreported recurring obligations.
Power Couple
Aiva is a predictive AI tool, using mature Large Language Models and proprietary capabilites making it more scalable and cost effective for large-volume processing than generative AI tools like ChatGPT, which can have runaway costs at scale. By using predictive AI and using some patent-pending capabilities that we’ve developed at Dark Matter, we can do that processing at scale, with accuracy at a much lower cost, meaning savings for our lenders.
Coupling Aiva with Empower, lenders can scale operations without adding personnel. In fact, our own human review needs have been reduced by 77% through Aiva automation. Working hand-in-hand with Empower, Aiva presents lenders with actionable data organically integrated with the loan process and clears the desk of routine work so teams can focus on exceptions, problem-solving and building relationships. The big picture stuff.
Sure, you could use just Aiva, or just the Empower LOS. However, they were created to use together to alleviate the tedium of document processing, so your team processes loans faster and with more accuracy.
Craig Rebmann is Managing Director/Product Evangelist at Dark Matter Technologies and has played a critical role in helping transform Empower into the most advanced and one of the most-used core origination technologies, now used by more than 80 lenders.