0
0
Few would argue closed loan acquisition is complex. Managing risk against faster turn-times, while trying to maintain relationships with sellers surely keeps many of correspondent investors awake at night. Go too fast, increase your risk. Go too slow, stress margins and ultimately lose sellers.
How can correspondent investors minimize portfolio risk while going for speed? By bringing intelligent, automated data and document processing technology together with rules-based automation to surface the facts of the loan, before due diligence begins.
But much of the industry’s data and document processing solutions sacrifice accuracy or cost to achieve speed. This manifests when lenders fall prey to using OCR alone, trust their LOS as the single source of truth and fail to perform automated analysis on the extracted data, increasing their risks and costs.
Let’s investigate each of these further.
Even the most sophisticated OCR is an island
Sure, OCR for mortgage data and doc processing is automated…to a degree. OCR at its core turns images into text, but that alone doesn’t determine the nature of the document. More sophisticated zonal OCR relies on templates to pull data out of standardized, structured documents (think Loan Estimate or Closing Disclosure). Unfortunately, this type of technology does not do well with unstructured documents (think gift letters). It would be virtually impossible to know where the needed information might be found in an unstructured document. There are even challenges with structured documents when information changes position across multiple instances of that form.
While there have been improvements in this technology over the years, the value of it for the mortgage industry is only as good as the domain expertise a given vendor might have. Knowledge of the variety of documents found in a loan file and pre-built templates for those that are standardized is a must but falls short when variability is high. Add in poor resolution quality, loan file pages out of order, upside down and missing docs and you quickly realize OCR is not enough.
The LOS is not the single source of truth
Relying on a single source of data is the fastest way to sabotage accuracy. LOS data is often the go-to source to save time, but LOS data is only as good as a given sellers process to ensure it is updated with what appears on the final loan documents. And even with a good process, if some of that data is input by humans… humans make mistakes.
In order to determine the truth in seller packages and ultimately minimize risk, correspondent investors must incorporate data from a mix of sources. These include LOS data, DMS documents and smart documents in a variety of formats and other 3rd party data sources like credit reports from the bureaus. All these are needed for a 360-degree view of a loan file. LOS data alone can’t support true due diligence and the transparency investors need to manage their risk.
The weakest link in due diligence is human capital
In a stack of loans, do you know if the closing disclosure (CD) was dated correctly? Does the loan estimate match the CD? Did both borrowers sign the application? There’s no way to know without a set of comprehensive rules to reveal data errors, inconsistencies or missing critical documents prior to a prefund review. These rules are often manual checklists auditors spend an exorbitant amount of time sifting through to get to the facts, subject to human error and reviewer variability along the way. These costs and inefficiencies negatively impact investor margins and seller relationships.
Thankfully, Industry solutions are available and improving. Advanced machine learning has vastly enhanced document recognition beyond OCR alone. Using algorithms to cluster and classify documents speeds up cycle time and exponentially expands the types of documents that can be classified. Automated data extraction (ADE) has advanced to accurately pull information from structured and unstructured documents using textual analysis, not keywords, extracting anywhere from a few to hundreds of data elements from a document- template free. Customized business rules automation is combined to create purified, factual data for correspondent lenders to complete pre-purchase reviews in less time with higher levels of confidence.
