HousingWire sat down with Phil Huff, SVP of Mortgage and
Real Estate Services at Altisource, to discuss artificial intelligence and
natural language processing in the mortgage space. Huff has been in the
mortgage industry for more than 20 years, serving as president and CEO of eLynx
and then Platinum Data Solutions. In 2017, Huff joined Altisource where he now
runs Mortgage & Real Estate Solutions, which encompasses five business
units, including Trelix Mortgage Fulfillment, CastleLine, Springhouse Valuations,
Premium Title and Settlement Services and Granite Construction Risk Management.
kind of changes have you seen in the mortgage industry over the last few years
when it comes to using artificial intelligence (AI) and natural language
Huff: The last few years is when the use of artificial
intelligence (AI) and machine learning (ML) technology has come into its own;
it’s not just an idea anymore.
Everyone always says the mortgage industry’s a bit late to
the party when it comes to implementing technologies like these, and I think that’s
been true in the past. But with AI and machine learning, it’s real today. It’s
what’s being used, and that’s the biggest transition we’ve seen over the last
HW: In your
opinion, what are the best applications of these technologies in the mortgage
Huff: The two main categories, as I see it, are the
front-end point of sale (POS) and the back office, which is managing business
processes, which is what we do a lot of here at Altisource.
On the POS side, you’ve got a myriad of checklists and
questions that need to be answered prior to getting the home buyer in front of
the right product. That can be streamlined through the use of AI and bots on
the front end, and there are a lot of different applications in our space for
how that’s being accomplished.
Specific applications in back-office automation involve the
use of neuro-linguistic programming (NLP) to augment optical character
recognition (OCR), the use of machine learning in image recognition to automate
some of the assessment of photos and images in the appraisal process, and the
use of machine learning in automated valuation model (AVM) technologies.
HW: How does
Altisource use AI to improve the accuracy of your AVM applications?
Huff: We use AVMs in a few different applications: on
the originations side default side and rental side. We use AVMs to estimate
sale and rental property values, analyzing property characteristics, the
neighborhood characteristics, recent sale listing data and recent market
transactions. We use custom machine learning and numerous proprietary data
sources to create property rental estimates. Combining this technology and data
enables us to get better over time, which is always the goal.
Our RentRange product utilizes AI/machine learning methods to “learn” how to drive more accurate property rental estimates. We actually have both pieces of the transaction – we have the upfront estimate and then the actual rental value, post-transaction, which we can compare to. It’s the perfect application for getting better at estimating property values.
We also have a sale AVM that we use internally, our Edge
product, which uses custom machine learning models for value estimates, and
it’s our own proprietary tools that we’re using behind the scenes.
Hubzu, our online real estate marketing platform,
incorporates ML technologies to assist in our initial estimate of value. In
this way, we get better at estimating the right value to place on a property
when starting the bidding process.
heard a lot about AI and machine learning but maybe not as much about NLP. How
does Altisource leverage NLP in its solutions?
Huff: When you augment OCR with NLP technology, we
now have the ability to recognize not only letters but the entire context of the
words in a document – not only bits and bytes but the words and sentences in
the structure of a document and even what kind of document we’re handling. We
can introduce intelligence on top of the OCR process, and that’s really the
advantage of NLP technology for us here at Altisource. We use it in our Trelixbusiness, for example. We use it in title services; we use it when we’re
listing properties. At a high level, it enables our systems to understand the
context of data in a document, not just a character in that document.
HW: What are some other other areas where Altisource is implementing AI?
Huff: I think the use of AI technologies in image
analysis holds a lot of promise, and it’s very, very exciting.
Using AI, we now have the ability to perform search and
match techniques on images or photos. Think of it as dragging a photo into your
search bar and hitting enter. In a split second, you could determine if a photo
is original or if it was actually a copy of something else that exists out on
Some other powerful applications of that image-match
technology are driving some early-stage fraud detection, some early-stage image
matching and the use of photos, knowing what image goes with what property address.
Think about being able to recognize if a picture contains an image of a one-story
versus a two-story home. Even looking to drive the back office of the equation –
looking at checks, for example, and being able to pull amounts off the check
and look at the writer/signer of a check and being able to analyze the image
from that standpoint as well. Marrying all these technologies together, the
possibilities are limitless. We just need to focus on the areas where we get
the most value for our consumer. I think this holds a lot of promise for the
of the future, how do you see the mortgage industry further adopting
technologies like NLP and AI in general?
Huff: If you think about it from the POS side,
that’ll continue to improve and get better. I think there’s still an
opportunity for it to be adopted more broadly.
On the back end, I think companies like Altisource are
continuing to lead the way from a technology perspective. Our Trelix solution,
for example uses technologies in the back office to automate a lot of the
routine processes to make it easier to integrate with existing systems.
But there are a lot of technologies that are going to be
used by everyone. It’s not as though this is an expensive path anymore. A lot
of these technologies are available off-the-shelf, even in open-source form
that companies can begin to do testing with. I think we’re going to see the
ramp of this technology much more quickly than we saw, let’s say, the ramp of e-mortgages
or extensible markup language (XML), or even before that, electronic data
interchange (EDI) and the web. I think it’s going to happen a lot faster and
the adoption curve will be much steeper with this technology than in other
For more information on Altisource solutions, visit here.
The post Executive Conversation: Altisource’s Phil Huff on AI and NLP in the mortgage industry appeared first on HousingWire.