00:06From Las Vegas, Nevada, I'm Allison LaForgia, managing editor of HousingWire's content studio,
00:12and today I am sitting with Max Klein, the founder and CEO of Lonelight. Max, thank you so much for
00:18joining me today. Thanks for having me. So let's start a little bit with the vision behind Lonelight.
00:23What opportunity did you see in the non-agency market that led you to build it?
00:28So the non-agency space has gone through what I believe like a tremendous transformation from
00:35a bit of an outside misnomer to mainstream, one of the critical drivers of the modern
00:40mortgage landscape. And with non-QM specifically, I think the challenges have become so apparent
00:48and combined with the growth and interest on the lender side that new tooling was
00:55critically needed to drive innovation and to create this, to transform this into a scalable,
01:01verifiable asset class akin to agency lending. And so when we were developing the tooling around
01:09Lonelight, we really wanted to focus on non-agency because we thought there was obviously a tremendous
01:14opportunity. But secondly, we didn't see many players in the space, unlike agency technologists
01:20who were focusing on like the bigger market. We saw this as, you know, where the market was going.
01:25And so we, we wanted to build tools that we essentially were skating to where the puck was
01:31going instead of where the puck was. And so Lonelight was born from the need to better serve
01:37the growing, uh, the growing economy of non-W2 employees. Um, it was built to serve lenders
01:43who want to serve these borrowers, um, and really just maximize the impact that an underwriter can
01:48actually have on the decision-making process, um, instead of spending time pouring over documents
01:53and, you know, staring and comparing. To your point, non-QM operates without the standardized
01:59system that exists in agency lending. Yeah. How does Lonelight approach that challenge differently?
02:06Yeah, that's a great question. So unlike agency lending, which, uh, relies on the backbone of
02:12DU and LPA for decisioning, non-QM has no equivalency. And so what we wanted to build at Lonelight was
02:19a intelligence layer that standardizes, uh, quality, eligibility, and liquidity, liquidity.
02:26Uh, we're focusing on the quality and eligibility layer first, but we wanted to create a system that,
02:34uh, that can provide both lenders and buyers, kind of the whole ecosystem of non-QM, uh, a shared
02:42underwriting language that they can rely on. Um, and so we're starting with the quality and
02:48eligibility layer to ensure that when lenders receive files from, uh, brokers or third-party
02:52originators, um, they're clean, preconditioned, ready to go. Um, and for aggregators that rely on,
03:00you know, the purchasing of these, of these loans, um, we can provide them with greater certainty
03:06downstream by helping the lenders upstream. Now, in your first answer, you talked a little bit
03:12about underwriters. Where do you see automation delivering measurable value and reducing friction
03:19specifically for that underwriting process? Yeah. So we have a belief at Lonelight that we are not
03:25replacing the underwriter. We think the underwriter still plays a critical role in this process.
03:30You know, from the domain expertise to ultimately like they, they are the arbiters of the decision.
03:35We just believe that they spend way too much time on, uh, inefficient work that honestly is best
03:43served by machine. And so we took a look at the process and we came to the ultimate decision that,
03:50what would an underwriter's life look like if they didn't have to look at documents, that they had
03:55a single pane of glass, that they had all of the information needed to make a decision. So what we're
03:59trying to do is empower the underwriters, um, by better augmenting their workflow and accelerating
04:05the path to decisioning. But ultimately they're the ones who make the decision just based off of
04:11cleaner, more readily available data. So Max, I have to dig in a little bit further there.
04:17I need a little bit more clarification around how Lonelight approaches the balance between AI tools
04:22that we're talking about here and that underwriter expertise that you just mentioned.
04:27Yeah. So at Lonelight, like we specifically built our tooling to support the underwriter decisioning.
04:35And so one of the things that we've determined through, you know, spending years in the space
04:40and really understanding the, uh, these AI tools that, you know, a lot of folks are, are building
04:45against is, you know, I personally don't believe that a full automated AI system is ready for
04:51a mortgage underwrite. Um, I think potentially it's best served in, you know, uh, smaller assets
04:58where there's kind of less liability, not a $500,000 mortgage. And so, um, what we're, what we're
05:06building is really the tool set to provide underwriters, uh, the pathway to make a better decision.
05:11So our tools are all about extracting information from documents, synthesizing them, running them
05:17against rules, presenting facts and findings to the underwriter. And then the underwriter takes
05:22that information and then ultimately moves the ball forward to the decision. Um, but again, we are
05:28building our solution around the underwriter knowledge and an underwriter's workflow. But what
05:33we're doing is again, really accelerating the path, the we're accelerating their path to make a decision
05:40based on the data that we're pulling from the source material, from the documents and the data
05:46in the loan origination system. Now there is a lot of hype in the industry right now about AI and
05:51lending,
05:51but looking at what's actually projection ready solutions require accuracy, traceability, and
05:59so importantly, compliance. What do you think the industry misunderstands about tools that actually
06:05drive value and meet those conditions? You know, this is really interesting because I see it all the
06:10time on, you know, everybody sees it on LinkedIn and things like that. It's like, you know, oh,
06:15I built this in a couple of days. And I think there's a couple of challenges there. I think number
06:20one, I think a lot of lenders in the space, they look at all of these posts, they look at
06:24the
06:24developments around open AI and anthropic, and they ask themselves, Hey, can I do this? Why don't
06:30I just do this? And so I think the build versus buy question is becoming much more apparent. But I
06:35think the challenges and you know, challenge that we face bringing our products to production is you
06:41could build something really slick in a couple of days. But ultimately, when rubber meets the road,
06:46when you actually put your product into production, everything breaks and every technologist who's built
06:51these products, they know it too. And so I think you know, we need a real focus on how to
06:57take these
06:58products from, you know, building demos on structured standard data, and productionizing them in the messy
07:06world specifically around, you know, non agency. And so we spent months with several design partners,
07:14building out our tooling, working with production data to ensure that, you know, when we made the leap
07:19from testing to production, we had a lot of our bases covered, you know, we still found, you know,
07:25really interesting opportunities to obviously enhance the tool set. But working with our design
07:30partners has really helped kind of accelerate our path to building a true production ready system.
07:34Now, Max, I'm going to ask you to put on your forward looking lenses here, and look at how you
07:42see AI
07:43shaping the future of non agency lending, underwriting, and lending operations.
07:50So I see, I see, I think there's a lot of I think AI is like, it's such a broad
07:58term, right? I think
07:59ultimately, what, what a lot of lender what a lot of folks in the ecosystem want is they want better
08:04tools to help them do their work better, right? Whether that's, you know, AI, whether that's automation,
08:08whether that's good old fashioned software. And I think one of the things that we're going to be
08:12seeing and stuff that we're actively looking at every single day is, what are the things that,
08:18you know, AI and LLMs and agentic workflows are really, really good at? And what are the things that,
08:23you know, sometimes old fashioned software can provide just as good? Because one of the challenges
08:29that I see in taking these tools to the mass market is these are non deterministic tools, right?
08:35You know, you give an LLM an appraisal, and you'll get probably three different answers with,
08:41you know, the same question. And so I think what's needed is to really understand where these
08:48tools fit into the various workflows. Again, I think taking an LLM or an AI agentic system from
08:58submission to decisioning, I don't know if we're ready yet. I think we will be getting very close,
09:04slow. I think as more, you know, as the builders of these tools, they ingest more production data,
09:10they see more loans, they train their models better. I think a lot of like domain specific
09:14models are going to be introduced. We're working on one as we speak, actually. And I think,
09:20you know, we're going to see a transformation from, you know, humans doing, you know, 90% of
09:26of the job to humans doing the last mile, right? And who knows, you know, five, 10 years, I think
09:34you could very well see, you know, full end to end automation, you know, click a button to apply,
09:39and you know, within 24 hours, you could get a mortgage. But I still think we're quite a ways from
09:44that.
09:45I mean, the pace of innovation over this past two years with the introduction of generative AI and AI
09:51tools is just absolutely crazy. So Max, thank you so much for joining me today. Thank you. I cannot
09:57wait to see what's next for Lonely. I appreciate it. Thank you so much.
Comments