Coffee Talk: How Will AI Change Our Industry? NTC’s Checklist for Evaluating New Technology

NTC_Coffeetalk_Webpage graphic

In an industry known for its acronyms, probably no acronym gets bantered around more these days than AI. Will it, as advocates predict, revolutionize the way we finance and support homeownership? Or is our industry years away from the practical adoption of AI, as some skeptics say?

In this issue of Coffee Talk, Jimmy Walby, Executive Vice President of NTC Operations, and Danny Byrnes, Chief Revenue Officer of Covius, share their views on AI and how NTC may deploy this game changing technology in the future.

Let’s start at the 50,000 ft level: where across the mortgage spectrum do you see AI gaining the most traction?

Jimmy:  First, we have to differentiate between the hottest new AI technologies, like generative AI, which has been getting most of the headlines recently and some iterations or subsets of AI, like advanced Optical Character Recognition (OCR) and large language models (LLM). OCR and LLMs have been around longer, are continually getting better and are already being used in both origination and servicing.

As costs to service loans increase, many servicers are employing chatbots to communicate with their borrowers and provide more on-demand support. This will only increase in the future given the industry’s need to reduce costs and consumers’ desire to self-serve and not talk to humans unless absolutely necessary.

OCR is being used selectively in both origination and servicing, and while it is getting better, it still has limitations. It’s great, for example, if you have standardized forms where the data is always in the same fields.

Where it tends to do less well is when you have a variety of documents and these documents have stamps, handwritten signatures, writing over names, etc. – in short, your average mortgage or title packages. OCR is much improved, and it is slowly going to defeat these challenges, but at least for the near future these files are going to need to be touched by humans.

How does NTC evaluate new technology?

Jimmy: We’re looking at new technology all of the time, and we get great input and guidance from Covius’ IT team on the value and soundness of the solutions we’re considering. But before we make a decision to implement a technology, it has to check four boxes for us.

  1. Does it work?
  2. How accurate is it?
  3. Is it cost-effective?
  4. Does our partnership with the developer have any strategic advantages?

At the moment, AI and advanced OCR are coming closer to checking boxes 1 and 2 for us. But at the end of the day, our core businesses—lien release and assignment of mortgage—require us to deal with high volumes of unstandardized, unsorted documents that need to be separated, recognized and prioritized. Today, that still requires a reviewer who can quickly discern what mortgage related documents need to be looked at to support the document being prepared.

Our current technology platform and vendor supply chain are extraordinarily accurate and efficient. Will AI catch up? Perhaps. More likely it will make our reviews even faster and more efficient rather than replace them entirely.

Having said that, I do believe that over time, perhaps in several years, technology will get us to the point where perhaps as many as 80% of our files can be reviewed without any human touch. A good analogy, I believe, is what’s happening with eRecording. Fifteen years ago, only 5% of our recordings were done this way, now it is 90%.

Danny, is this the way you see it?

Danny: Jimmy is never wrong from an operational perspective. From where I sit in business development, I see other potential uses for AI.

Consider a use case like this – a client is cleaning up an acquired pool of loans, to avoid downstream issues or to prepare for sale. In the past, they would have to rely on the accuracy of the data in their system, provided when they acquired the loans or from a custodial exception report. I don’t think it is any secret that the accuracy of these various data sources is questionable at best.

With technology now, loan image packs can be processed (going through hundreds of documents, correspondence, etc., per loan) to quickly identify missing title policies, mortgages, assignments, notes and other pertinent mortgage-related documents. This technology would enable us as a vendor to provide a quick and accurate estimate of what it will cost to remediate these issues: Is it $15,000 or $150,000 or $1.5 million?

The next thing I see coming is logic to not only identify the documents but also make determinations, similar to what Jimmy discussed earlier, validating the efficacy of a document, making decisions on the best path forward for remediation and even kicking off tasks for downstream workflow. For example, there are five loan modification documents in the file – which one stuck? Or a promissory note is in the file, but it is not for that loan. Is the loan that note is for in this pool, and is that file missing the original note? Title policy present? If so, is it the right one? Is there a break in the assignment chain? Are errors on the document identified as simple scrivener error or critical? There is a very long list that would benefit the process, save time/money and cause more adoption of the technology.

Jimmy, you had an interesting observation about timing. Maybe we could end with it?

Jimmy: Some of the biggest breakthroughs in AI happened in 2023, which was a challenging year for the mortgage industry. Volumes were down; profitability was squeezed, and few investments were being made in technology.

If generative AI had been more available in 2018, the mortgage industry would have had the interest, the bandwidth and capital to invest, and it would have saved servicers and their vendors millions during the record volume years from 2020-2022.

Share This Post

Scroll to Top