- 5 months ago
- #aidatascience
- #mortgagetech
- #realestateai
AI in mortgage isn’t being held back by ambition; it’s being held back by infrastructure. Amy Gromowski, VP and Head of Data Science at Cotality, sits down with HousingWire’s Allison LaForgia to break down why becoming AI-First starts with one thing: data that’s ready to be used.
#AIDataScience #MortgageTech #RealEstateAI
#AIDataScience #MortgageTech #RealEstateAI
Category
🗞
NewsTranscript
00:06I'm Alison LaForgia, and for this conversation, I'm joined by Amy Gromowski, the VP and Head
00:14of Data Science at Potality. Thank you for having me. Thank you for joining me. So Amy,
00:20to jump right into the deep end, we hear the term AI ready thrown around a lot in the mortgage
00:27space. But many in our audience are still struggling with silos. How do all of the players
00:34in the built environment move the industry forward to be AI first? So AI has opened up a world of
00:42possibilities for everything we do. I use it for legal research, medical research. We know there's
00:51a ton of value in individual productivity. Maybe you're using it to develop the outline of
00:57a paper you're writing or an article, whatever that is. There's an individual personal productivity
01:03play. And then now we're in already the era where there's one company, I think maybe there's more
01:11now, but the last I saw, our first billion-dollar revenue company with two people. So the opportunity
01:18for AI is small and it's large. So I think that leaves us all feeling a little bit like, where
01:25do we start? There's a short-term, from where I sit, there's a short-term play and there's a long
01:31-term
01:31play. And if we just start focusing on the short-term play, focusing on individual productivity
01:39within a company, within whatever it is that you're trying to do, get familiar with the AI,
01:45and then move to enhancing and starting to streamline some of the existing work that you do.
01:52Maybe it's through partnerships. Maybe it's through your own AI capabilities, such as enhancing user
01:58experience and enhancing customer experience using AI and large language models, all the way to this
02:05idea that we can reimagine mortgage. We can reimagine real estate. But I think in order to get to this
02:11real reimagined state, we need to take this in smaller increments and really get to a place
02:17where we have our own understanding of what's possible and how to get there. And once we have
02:22that, through partnerships, through a lot of relationships in terms of what do we want the
02:27future to look like, we can really start building towards a future state, whether that's agents to
02:33agents talking. But getting to that point, it starts with short-term, medium-term, and then we can get to
02:39the long-term. The implementation of AI in mortgage, outside of mortgage, across our lives, has really
02:48changed the way we interact with technology. And we've seen a shift in mortgage and in other businesses
02:56from traditional dashboards to natural language interfaces. From a data science perspective,
03:03what are the challenges of making talking to your data actually work across different personas from
03:10the C-suite to the loan officer? Yeah. And the persona piece is big, a big deal. So natural language
03:19and AI, it changes the game, right? The way that we have interacted with data and analytics has been
03:28either through UIs or APIs or bulk data transfers. But this idea now that, and that's very limiting.
03:36It's very limiting in terms of, you know, let's say a UI, it's point and click, and you have to
03:42have
03:42some pre-canned analytics. I have to imagine what it is that you want to learn from Cotality
03:49and have it designed in that way. So it's fairly rigid. With natural language, it completely opens up
03:54the realm of possibilities and breaks down that structure, which is super, very exciting. But when
04:02you're a company like Cotality that has 16 petabytes of data, five and a half, you know, billion records
04:08that we're managing with our aerial imagery partnership with something like 38 billion square
04:15meters of the U.S. that we're processing, right? So that's a massive amount of data. And it covers
04:20a wide range of domains from understanding specifics about a property, you know, maybe you're looking
04:27for a specific property characteristics in a certain valuation or, you know, price range, and you want
04:34to talk to data on very straightforward, you know, ideas like that. That's a little bit more easy,
04:42but when you have to accomplish, right, with a data agent. But when you have so many domains that can
04:48cost hazard, that can cost geospatial aspects. So now let's say I want to ask a question like,
04:54you know, what are the homes that are available in the Denver area with low wildfire risk between
05:01$500,000 and $600,000? You have now crossed a lot of data domains. I have to define Denver. I
05:07have to
05:08look up valuation information. I have to understand hazard risk. I have to traverse all of that data and
05:14bring that back into a high quality response for a user, um, whoever's interacting with our data.
05:22So that, um, that's a level of complexity in order to deliver for quality and at scale that is,
05:30I think, unique in a way to quotality versus maybe some other data providers or, um, companies that can
05:37build one agent and talk to a data. But that is what we're working on. And we're moving into,
05:42into a place where, um, you, all the capabilities around, uh, interpreting the question with semantic,
05:50semantic layer, building out knowledge graphs and how the data, um, talks to each other and the
05:55relationships there, and then how to build queries that's at scale to traverse all of those different
06:01data assets. It sounds like there is so much going on. And I love to hear your perspective for
06:10a lender or a real estate platform looking to scale when they're looking to scale time to value really
06:18is everything. How does quotalities approach to AI first about data delivery change the timeline
06:25for a company looking to deploy? Maybe it's first truly intelligent AI agent.
06:32Yeah. We are all about meeting the market where they are. We're all about, you know,
06:37where our clients, where are our clients and their users doing their work? Where does our data need to
06:42be? So whether that, you know, today that's bulk data delivery through secured FTP. Um, it's a,
06:49we're available in the cloud marketplaces like Snowflake and Databricks. Uh, we're available
06:54via API if you want to engage that way or UIs. And I was just talking about, you know,
06:59point and click and getting data that you want. And we're moving now to AI ready data and MCP servers.
07:06So this is, you know, AI ready data is these companion files that you deliver along with bulk
07:12data that gives the instructions and how to read our data through natural language or MCP servers are
07:19the, the wrappers around API that can use natural language to traverse those APIs. So that's the journey
07:26that we're on. And I mentioned all the different data domains that we have. So many, right. It'll
07:31take us a little bit of time, but we're building out that factory and, um, you know, listening to the
07:36market in terms of where, where are our customers going, where their customers going. And then
07:43ultimately this opens up the space for, you know, agent to agent. There's just, there's a lot in
07:48the mortgage and real estate space. There's a lot of documents, right? And that is really
07:52ripe for, for AI so much efficiency to be gained there.
07:57It sounds like there's so much data that Cotality has to work with to really build out these AI
08:02capabilities. And it seems like you guys are not only well into this journey, but there's so much
08:08future capability that you're looking to build out. And there's so much research that you're
08:13continuing to do in the space to further support these efforts. And one bit of research that you've
08:19done recently is into consumer sentiment. What have you discovered about the breaking point
08:26as far as trust when AI starts advising a home buyer at a critical moment?
08:33So the, the recent, um, consumer survey that we did, which was our second one,
08:40it was really interesting to me because the, the question around trust in AI and the home buying
08:46process went from one in three people. So 33% said, yeah, I would trust AI in the home buying
08:51process to 16%. So it was cut in half in less than a year between these two survey periods,
08:59right? Very interesting. And so we think about why, why might that be? And, and we didn't really dig into
09:06the why, but if, if you just take a step back and you think about the journey that every one
09:10of us have
09:11been on that I just mentioned earlier over the last year or so, people are coming to, um, the realization
09:18at scale here that AI is powerful and the possibilities are, are really endless. So what
09:26does that mean in the home buying process? Does that mean that, um, you know, in the real estate
09:32transaction process, in the, in the mortgage underwriting process, that is highly personal,
09:37um, in terms of, it's a, it's a big financial transaction for the end consumer. It's an
09:43emotional typically, right? Transaction for the end consumer. So as you think about that and you
09:48think about AI, what does that look like in terms of knowing what's happening, trust and quality in
09:54the process and having the ability to be part of it? If you are the consumer, I want to know
10:01more
10:01about, you know, what that AI has advised, let's say, you know, in terms of the loan or the interest
10:08rate. So, um, I think it's really important that the consumer is part of the journey that we're
10:15listening to the consumer and able to provide a level of transparency and quality, uh, to ensure
10:22that the adoption is there, that the trust is there. To that point, there is a lot of anxiety about
10:29AI replacing the human touch in real estate. And you just mentioned that this is a massive
10:34transaction that can be very emotional and anxiety prone as for the end consumer, as you're going
10:41through it, you've advocated for people supported by AI. What does that look like in practice during
10:49those high stakes moments of a transaction? Yeah. The property, the, the residential property
10:55industry in the U S is worth, um, 55 trillion. It's just looking at that. I think it's up from
11:0343
11:03trillion recently. And the average house price or median is 460,000, 461,000. So imagine for an
11:13individual, you know, what that means in terms of their long-term financial commitment, where they
11:20want to live, potentially raise their family or retire, whatever that looks like. So who are you
11:25going to trust in that process? Who are you going to look to? We are still people. And we look
11:31to
11:31people, right? Real estate agents, they want to build relationships and, um, you know, meet new
11:38people and have their own personal style and brand, be part of what they offer and individuals or
11:45families or, you know, whatever you're buying, even investors, they want to trust somebody who knows a
11:50local area who understands what they're looking for and guide them through a process. So where AI,
11:56in my opinion, comes to play is all that administrative work, let people do what they do
12:02best, which is engage with people and bring their expertise to the table in the important discussions
12:08and transactions and the time that they spend with other people, filling out paperwork, analyzing
12:14documents, um, and even just through the, the loan process, how much back and forth and analysis and
12:20tracking down of additional documentation is required. That all puts friction points on people. So if the AI
12:26just starts tackling, you know, administrative type work, then that really frees up people to guide people
12:33through the process and builds trust.
12:36And giving that reassurance during that really emotional transaction, I think that that's such a
12:44fantastic point, which brings me to where I want to end for today before we get into too much more
12:52by talking about the consumer, the end user in this case, when they're making the biggest purchase of
12:58their life, a black box AI tool just isn't going to cut it for that trust piece that we just
13:05mentioned.
13:06How is your team ensuring that the AI driving these insights is transparent and explainable to the
13:12end user?
13:13So responsible AI is king in my world. What does that mean? Responsible AI, data privacy and security,
13:21transparency in terms of being able to trace every, uh, step that an agent or AI is doing. What data,
13:29what's the question? Is it an appropriate question? What data is it tapping into? What's the reasoning
13:35going in behind, you know, what the agent is doing and having that type of visibility through the entire
13:42process and including the answer, um, and, and how it came, came to that. There's a, uh, you know,
13:51a process when in the development process around evaluation frameworks and metrics. So just following a
13:58very rigorous set of practices and principles around transparency around evaluation. And once those models are in
14:06production and ongoing surveillance and understanding of how they're performing, you know, and there's multiple aspects to that.
14:13Um, so, so that's around just understanding how the models perform, but there's also,
14:18you know, ethical guidelines and moral principles. So, you know, we have a whole AI governance committee
14:25and council that is made up of, um, cross-functional, uh, leadership. So not just our business segments,
14:33but HR communications, legal, outside legal council compliance, you know, monitoring what's happening in
14:40the regulatory environment, ensuring we're compliant. They're compliant with our client contracts.
14:45There's a lot of aspects to responsible AI and ensuring that you have a, a robust, um, set of guidelines
14:52and principles and principles and processes to ensure that every single one of your models are in a known
14:57state and have gone through that process. That that's the, that's what we do at Cotality that gives
15:02that, you know, stamp of trust to an end consumer or the property professional using our tools and our
15:09data that, um, you know, there's, there's a lot of oversight on, on the quality. Well, Amy,
15:17Amy, thank you so much for taking me through what Cotality is doing. You guys are deep into your AI
15:24journey. It sounds like there's so much more on the plate. There's so much more to come. And thank
15:30you for walking us through the responsible, transparent approach that you were taking.
15:34Yeah. Thank you, Alison. It was great to be here.