Smarter, faster, safer: Reimagining mortgage underwriting with AI

HousingWire

Artificial intelligence is transforming mortgage underwriting.  Join our webinar to get insights from industry leaders as they reveal how top lenders are using AI to streamline originations, reduce costs, and empower underwriters to take on a more strategic, value-driven role. Explore how with AI-powered workflows handling high-volume, low-risk applications, underwriters will be free to focus on complex scenarios that demand human expertise and judgment.  The future of underwriting is forward-thinking teams redefining the underwriter as a modern risk strategist—combining critical thinking, pattern recognition, and customer empathy to make smarter decisions. All of this powered by AI will enable real-time, adaptive compliance, helping lenders stay ahead of regulatory demands and manage risk seamlessly across  regions.

Transcript

00:00:00Good afternoon, everyone, and welcome to today's webinar.

00:00:06I'm Alison LaForgia, Managing Editor of the Content Studio at HousingWire, and today our

00:00:11topic is Smarter, Faster, Safer, Reimagining Mortgage Underwriting with AI, presented by

00:00:18Sutherland Global.

00:00:20Now, before we dive in, I want to start with a few quick housekeeping notes.

00:00:23If you have any questions that you would like to send to the panel, you can drop them in

00:00:28the Q&A, and we will make sure that Sutherland's team gets back to you.

00:00:32Today, I'm really excited to jump into this content.

00:00:36It is a really exciting flow, but I want to start with introducing our subject matter experts

00:00:41who are joining us today.

00:00:43So let's start with Randall.

00:00:46Randall, can you please introduce yourself?

00:00:49Yeah, absolutely.

00:00:50Hi, I'm Randall Garland.

00:00:52I am SVP, Head of Mortgage Credit for Next Bank out of Dallas, Texas.

00:00:57Celebrating my 30th year of mortgage credit this year.

00:01:02So lots and lots of time on the desk, underwriting.

00:01:06Hope to bring as much information as possible to this call and to this webinar.

00:01:11Thanks, Randall.

00:01:12And now let's jump over to Joy.

00:01:14Hi, everyone.

00:01:15I'm Joy Simenskis.

00:01:16I am the Director of Sales and Business Development for Calix Software, the OG and Mortgage LOS

00:01:23origination software.

00:01:25And now let's jump over to Hitesh.

00:01:30Hi, everyone.

00:01:31I'm Hitesh Karwa.

00:01:32I head the mortgage solution and transformation at Sutherland.

00:01:36Close to two decades in mortgage industry.

00:01:38That's what I did.

00:01:39Still loving the hustle.

00:01:40From a loan officer to a consultant to a digital practitioner.

00:01:44Happy to be here.

00:01:45Thank you, Hitesh.

00:01:46And now, last but not least, we have Andrew.

00:01:49Hey, everybody.

00:01:50Andrew DeGood.

00:01:52I'm the CEO of Ask Bob AI.

00:01:55We focus on knowledge agents that don't hallucinate and operate at a 99% plus accuracy rate.

00:02:04I spent about 20 years in the secondary market around mortgage-backed securities before transitioning

00:02:11to the wonderful world of tech and AI in 2022.

00:02:14And now, without further ado, I am going to pass things over to Hitesh to start with today's

00:02:21content.

00:02:21Hitesh, the floor is yours.

00:02:25Thank you so much.

00:02:26Before we deep dive into today's topic, I just want to give a brief overview about Sutherland

00:02:31mortgage.

00:02:32The lens through which we look at Sutherland, there are three layers, there are three horizons

00:02:37that we technically work on.

00:02:39The fungist and transformation, that's the core foundation to our business, which is our

00:02:44domain and our talent expertise of servicing originators, services, third-party reviewers

00:02:51for more than two decades now.

00:02:53We leverage that domain with Sutherland's enterprise-wise platforms, which are verticalized to provide

00:03:02operational efficiencies and reduce technical tech.

00:03:05And then we include the whole ecosystem where we partner with companies like Google, Microsoft,

00:03:11and more specific tech companies like CalyxPath, Joy, is one of our partners.

00:03:19We bring everyone together, the domain, Sutherland, TOOTS, and our partners together, and then design

00:03:25what the future transformation looks like.

00:03:29On a brief, that's what Sutherland does from domain and from transformation.

00:03:33We can now move to the topic of AI in mortgage origination.

00:03:44Now, as we are aware, the last few years have been, you know, we've seen lean inventory.

00:03:51Typically, everyone is chasing the same border.

00:03:55The competition has intensified, and the rise of fintechs are challenging legacy platforms and

00:04:01processes.

00:04:02But AI, in fact, has been there for quite some time now.

00:04:08If we go to the next slide, here are some data points.

00:04:15We've seen machine learning, you know, helping with some credit scoring models.

00:04:20The last five years, we've seen chatbots, a rise in chatbots for answering customer queries.

00:04:25In fact, three months back, Fannie Mae came out with a fraud, AI fraud solution.

00:04:35They partnered with the AI platform, which means that AI is definitely a game changer if you

00:04:42don't look at it as a magic.

00:04:44If you look at it as, I would say, a car that has cruise control in it, where you are at the

00:04:50driver's seat, it's not a self-driven car, then probably it's a game changer.

00:04:56I would, you know, probably turn this over to Andrew, who lives and reads AI, to get your

00:05:02thoughts on that, Andrew.

00:05:03Yeah, no, actually, I'm stealing that, Hitesh.

00:05:08That might be one of the best explanations I have actually heard when it comes to that.

00:05:13And I think that is the biggest thing that we're seeing right now.

00:05:16You know, I've just come out of a few different conferences speaking on AI and hearing others

00:05:21talk on it.

00:05:22And I think for right now, you know, the biggest piece of it is it's the lift, right?

00:05:28So the cruise control example is kind of perfect here.

00:05:33You know, we're seeing companies out there like, you know, a trained who's doing it, you

00:05:38know, an 80% lift, right?

00:05:40When it comes to document processing on the underwriting side, we're seeing companies that

00:05:45are doing voice coaching, you know, to assist, you know, their LOs and anyone else in their

00:05:54call centers, you know, seeing the advent of knowledge agents, which are allowing for

00:06:00information to be accessed at a much, much faster rate within the space.

00:06:06But ultimately, at the end of the day, we still have that human component, right?

00:06:10I think the doom of AI is going to take all of our jobs is very much an overhyped thing

00:06:17in the media.

00:06:18And the reality is, is this is a lift, not a replacement.

00:06:28Absolutely.

00:06:29Completely agree with you.

00:06:30I mean, it's going to augment the processes and, you know, provide that lift.

00:06:36I mean, it would be interesting to hear, you know, Randall from you from an ops

00:06:40perspective.

00:06:40I know technology adoption can be emotionally exhaustive with so much information flowing

00:06:47around on AI.

00:06:48There's a lot of pressure, AI pressure, I would say, from an ops perspective.

00:06:52So as a, you know, from a bank perspective, right, and you heading to the credit division,

00:06:57how do you look at AI?

00:07:00Yeah, I think, you know, I'm probably a little less conservative than some in the banking

00:07:06space.

00:07:06We obviously have compliance and QC that are involved with concerns about security and what

00:07:14AI will look like in the landscape of banking as it applies to compliance.

00:07:21From our world, I think, to your point, it's very much an assist type role for us, at least

00:07:29in this current landscape as we get our hands around what AI will look like.

00:07:34What is the security of AI?

00:07:35How does that grow our business?

00:07:38And what risk does that pose for us in the new space?

00:07:43Today, what we envision AI being is more, as you mentioned, an ability to scale without

00:07:50adding extra headcount to that volume, especially as we potentially look at a refi boom, if you

00:07:58will, as rates are proposed to come down even more through the rest of this year and into

00:08:04next year, we look at that to say, hey, what could we do different than we did in the previous

00:08:10refi booms?

00:08:11I think the equal for us is AI.

00:08:14If we could have a technology that indexes our files and grabs or collects the data, scrapes

00:08:20the data from what has been indexed and then starts to fill in the blanks that somebody

00:08:26manually would have to do, that certainly creates a lift and it creates an ability for us to

00:08:32scale our business.

00:08:34What happens when the technology can't read it or it's just a bad copy or on and on and

00:08:42on?

00:08:42Well, there's where the assist comes in.

00:08:44And so I don't, to your point, I don't think in the current environment it would be a complete

00:08:50elimination of humans, for sure.

00:08:53I definitely think that humans will have to have a touch in it at some point, at least

00:08:57in our current environment.

00:09:01Completely agree with that.

00:09:03I think if we're talking about AI, we have to talk about the tech stack on which it will

00:09:10be built on, right?

00:09:11The loan origination system.

00:09:14How do you look at it, Joy?

00:09:16Being a platform company, so many fintechs coming up with point solutions and, you know,

00:09:23having AI solutions being on top of LOS to enhance productivity.

00:09:27Absolutely.

00:09:28And we've been, this has been top of mind for five plus years.

00:09:32And we have native cloud solutions available, but the technology piece is so critical and

00:09:39building it out to a spec that makes sense is really the priority.

00:09:45And what I mean by that is we have companies who, and I think you mentioned it earlier,

00:09:50is it's not magic.

00:09:51It's not, you know, an instant fix.

00:09:54It takes time and you've got to build it out.

00:09:57Because what we found is as we offered this automation, some clients would be like, well,

00:10:01wait, we don't want that to be automated.

00:10:03We want that human element to pop in and prove things.

00:10:06And, you know, they want the efficiency, but not at the sacrifice of accuracy or compliance.

00:10:11So it is a phase where we're in the baby steps phase, in my opinion.

00:10:16I do not feel that we're going to eliminate humans, but I do think that there is a need

00:10:20to make things efficient, faster, and even more accurate and lower that risk.

00:10:26And that is at the forefront of our work right now.

00:10:30It's just continuing to build that out where it makes sense.

00:10:36Thank you for that point of view, Joy.

00:10:38Absolutely.

00:10:39I think, you know, we're seeing a lot of discussions with customers.

00:10:44And we always tell the customers that it's, I mean, you know, it's not going to completely

00:10:51overhaul your operation.

00:10:52There's no secret sauce.

00:10:56The secret sauce is how you approach it.

00:10:59The strategy that you apply to it, right?

00:11:02That's exactly what makes a difference.

00:11:03Because every organization, be it a bank or a non-lender, they have AI in some form or the

00:11:09other.

00:11:10But it's how you use it, where you use it, is going to give you the ROI.

00:11:17We can move to the next question.

00:11:22What does autonomous underwriting really mean?

00:11:28I think off lately, we've seen a lot of companies talk about different AI types, AI maturities,

00:11:37being an assistant AI from an assistant AI, helping with some content.

00:11:43Like, for example, you know, chatbots giving responses to humans who would review and then

00:11:49respond to the borrowers.

00:11:51You know, augmented AI, automated AI like RPAs, content summarization.

00:11:57And there are other AIs which technically pick up certain pieces and make the decision.

00:12:03How these things come together to help, you know, the entire lending lifecycle, be it

00:12:12fulfillment, be it servicing, is something which lenders and, you know, mortgage industry

00:12:18is looking at.

00:12:20So technically, you know, it's like a twin brother strategy.

00:12:25One brother looks at automation.

00:12:27The other one looks at the human, right, who looks at the expertise.

00:12:31And if I have to put this in the underwriting context, we look at it as, let's say, document

00:12:37classification that indexes the document, gets the data out, runs the rule, does the income

00:12:43calculation.

00:12:44And the judgment part is being left to the underwriter.

00:12:49So this 80-20 shift technically can start with 50-50 first to give underwriters some confidence

00:12:56that, hey, these non-core tasks that you're going through the piles of documents, you don't

00:13:01have to do that.

00:13:02There are certain automations that can take care of it.

00:13:05And eventually, it goes up to 80-20, where 80% of the work, the low-risk work, that the

00:13:11underwriter can, you know, keep it at the, for the AI to automate and focus only on, you

00:13:18know, rather than being a gatekeeper, we are strategic analyst, right, focus on the judgments,

00:13:24focus on the quality of loans, focus on fraud and other things.

00:13:27Is what we are seeing in the market and different AIs, platforms that we've gone through are

00:13:35coming up with this strategy.

00:13:38Any points, Andrew, how do you look at it, you know, being an AI organization, the different

00:13:44AI majorities in the market?

00:13:46Yeah, I think the biggest thing, and you just hit on it, is we're taking away, you know,

00:13:52slowly but surely those activities that just, quite frankly, didn't make sense.

00:13:57I actually started out as an underwriter a very, very long time ago.

00:14:01I still had hair, and files were paper.

00:14:05And so, you know, we would sit there, I feel like maybe Randall remembers this too, where

00:14:10we would do the stare and compare, right?

00:14:12Like, I remember being in, like, places and having a loan file sitting in my lap, right?

00:14:17And when I was doing all those extra activities that really took away from my bandwidth and my

00:14:23capacity to truly look at what was going on with the file and to truly understand where

00:14:29the risk was, right?

00:14:31So as we begin to remove those things off of the underwriter's plate, right, now their

00:14:38bandwidth, you know, because we only have so much bandwidth, each of our individuals, right?

00:14:43It's famously why Steve Jobs, you know, wore the same outfit every single day.

00:14:47And so I think what this is going to actually do is just increase the quality that we're

00:14:53getting from the underwriting side.

00:14:55I think also there's going to be an employee satisfaction component here, right?

00:14:59Because underwriters specifically are going to be able to focus on the things that they

00:15:04want to focus on.

00:15:05So maybe those cookie cutter loans, that might be something that we see that gets, you know,

00:15:11to a, you know, 90, 95% automated underwrite with a final checkmark, but I'm not spending

00:15:18time having to focus on that much more basic file.

00:15:23The other piece, and you touched on it, is the ability for employees to access information

00:15:28a lot more quickly is going to reduce the number of questions that are coming from other

00:15:35personnel within the organization to the underwriter, right?

00:15:39As we're seeing the emergence of underwriting scenario desks and AI, as an example, you

00:15:45no longer have context switching.

00:15:47There's actually a really interesting study out of the University of California, Irvine,

00:15:51that shows that the average time to refocus a knowledge worker when they've context switched

00:15:58is 23 minutes and 15 seconds.

00:16:01Now that's the average.

00:16:02But there's a significant cost to all of these originators when you're having to do all this

00:16:11context switching in and out.

00:16:13So I think you're going to see higher employee satisfaction.

00:16:16You're going to see, you know, reduced cost and just overall better quality.

00:16:22Absolutely.

00:16:26Now, no two loans are the same.

00:16:30And I'm sure, Randall, you'll agree with me on this, that even if the two colleagues who

00:16:35are working on two separate borrowers who are buying a property on the same street, it's

00:16:40a different loan altogether, right?

00:16:43The credit score might be different.

00:16:45Pay stop, the income, bank statement, asset, everything changes.

00:16:50So from an underwriting perspective, do you think underwriters will have that confidence,

00:16:56you know, where more than 50% of what they're doing currently has been automated and they

00:17:02can trust AI and the output that comes out of it?

00:17:07Yeah, totally.

00:17:08I think speaking to Andrew's point earlier, we did grow up in a time where paper files

00:17:16were the way that it happened.

00:17:17And I distinctly have a memory of going into one of our customer's offices with a, not

00:17:26now, this was about 20 years ago, but went into a customer's office with our sales associate

00:17:31and we were trying to get their business.

00:17:34We walk in and on the top of everybody's desk and processing was a light box.

00:17:40And they could turn these light boxes on, put down a signature, and then stick another piece

00:17:46of paper on top of it and perfectly forge the signature of that borrower or of another,

00:17:53off of a VOE or what have you.

00:17:57And so with that burned into my mind the concerns that we have about fraud and what misrepresentation

00:18:05looks like in our current environment, I use the saying that says desperate people do desperate

00:18:11things.

00:18:12So from an underwriter's perspective, I spend a lot of time really analyzing the document,

00:18:20looking for inconsistencies, looking for font changes and signature changes and things that

00:18:25are not consistent throughout the loan.

00:18:28Things that I don't have time to do as an underwriter, because we obviously have a lot

00:18:32of loans to get through, is I don't have the time to sit on a bank statement and take

00:18:37every debit and every credit and add and subtract them from the beginning balance to make sure

00:18:44that that ending balance really is accurate.

00:18:46Like we don't have the capacity to do that.

00:18:48So from an underwriter's perspective in a world where AI could take off of their conscience,

00:18:59I guess is probably the best way of saying it, that there could be fraud in this file and

00:19:04I need to evaluate every single document for inconsistencies and catch the misrepresentation

00:19:11would certainly provide a lift and certainly allow us to get through loans a lot more quickly.

00:19:18Yeah, absolutely.

00:19:21Since you touched on fraud, I just remember the CEO of Fannie Mae quoting that it took

00:19:28their fraud investors 60 days to detect fraud versus the platform detecting it in 10 seconds.

00:19:37And I'm quoting her, she's actually made that statement, which is like great.

00:19:44So I mean, especially in underwriting, if fraud is detected real time, if AI can check documents

00:19:53and say, you know what, the name is different.

00:19:56The borrower has submitted a pay stub, which has an entity where he's worked for two years.

00:20:01But when we do a LinkedIn search or a social media search, right, he's joined somewhere else

00:20:07like three months back, right, claiming to be a primary residence, but he already has another

00:20:13property somewhere, which has not been, you know, caught.

00:20:17And AI does that quickly.

00:20:18Of course, there are other sites to AI as well, where it can probably categorize borrowers based

00:20:27on the patterns that it has been trained on, which definitely, you know, that's where human

00:20:31comes in play.

00:20:32And, you know, you don't leave the decisions to AI.

00:20:36You let, you know, the human do those decisions.

00:20:40Joy, how do you see this playing out, this entire shift in AI maturity?

00:20:47Well, I will tell you, as we're kind of looking back a little bit and reflecting, and I started

00:20:52in 1998.

00:20:53I know that's been a long time, so I'm dating myself, but in long origination and then evolved

00:20:58from there.

00:20:58And I've seen the industry and the word evolution is really the best word because I remember

00:21:05those paper files.

00:21:06I remember the checklists on the front of the file.

00:21:09I remember exactly what we were talking about as far as the stories of people copying signatures

00:21:16and so forth.

00:21:17But I also have seen that same evolution in such a beneficial way when it comes to the

00:21:23sticky notes on the files.

00:21:24Now we have this, all these technology advantages where we're not using paper.

00:21:35We've gone to paperless.

00:21:36And now we've saved time in just handling these files and it's all electronic.

00:21:42It's all online now.

00:21:43And then we're moving to the cloud.

00:21:45And maybe there's a, instead of a manual sticky note somewhere that also compromises the integrity

00:21:52of the file or the privacy of the borrower, where it's all in your system.

00:21:58So it makes the review of the file cleaner, more efficient.

00:22:02And then as you automate things like underwriting conditions, automating fees, all of these things

00:22:08give that underwriter an advantage to skip past the mundane tasks and that Andrew was talking

00:22:16about and really focus in on the depth of the file.

00:22:19Things that AI can't necessarily see up front, the full picture.

00:22:23You know, maybe they've been on their job six months, but they didn't see that they were

00:22:28a student prior to that.

00:22:29And that may be a compensating factor under FHA.

00:22:31So there are a lot of details to a file that I'm not sure that we're ready for today with

00:22:37AI, but at least it can get through those tasks, the Monday check marks that take the

00:22:42majority of the time when underwriting a file.

00:22:45So I can see it really progressing.

00:22:48It has since I started in the industry and now looking forward where it's going.

00:22:52It is exciting.

00:22:54I think that we're going to be able to, these borrowers are going to be able to get their

00:22:57into their homes much faster.

00:23:00And I think that's the goal.

00:23:01And that's an exciting thing.

00:23:04Absolutely.

00:23:05I agree.

00:23:06Absolutely.

00:23:07I think, yes, I think before, you know, jumping into AI solutions for underwriting or maybe,

00:23:13you know, across the landscape, it's important to take a step back.

00:23:17And like you and Randall and Andrew mentioned, to take a step back, understand what problem

00:23:22are we trying to achieve?

00:23:24And then see, once it becomes very clear that, okay, we're trying to solve a document problem

00:23:29here.

00:23:30We're not getting clean data.

00:23:31Because without clean data, right, what models are we going to run on?

00:23:35So where do we need to intervene first, right, fix that issue, then move to another set?

00:23:45But just one question, Andrew, that I have.

00:23:47Do you think it has to be perfect before you run AI, right?

00:23:53Or should we, like, you know, pick up small pieces and, you know, try to master that before

00:23:59we...

00:24:00What are your thoughts on that?

00:24:01Yeah, so I think I have a couple of thoughts on it.

00:24:04Number one, I think it's definitely a crawl, walk, run situation.

00:24:08And I think Joy pointed out something really important that I love that you guys are doing

00:24:13over there, which is, you know, every mortgage shop is slightly different, right?

00:24:19And everybody is going to think about things a little bit of a different way.

00:24:22So the number one thing I would say is technology partners, when it comes to AI, we need to make

00:24:26sure that it's flexible and that we're willing to meet, you know, the customers where they're

00:24:32at today and what they want to do.

00:24:36I have lots of conversations with different lenders every single week, and I'm amazed at

00:24:41how big the spread is in terms of, you know, sort of how they're looking at it, right?

00:24:46So I think the accuracy level really depends on what are you trying to approach, right?

00:24:52Are you very early on and you're just using Clotter GPT to help, you know, people, you

00:24:59know, role play in a sales aspect?

00:25:03Are you looking to help, you know, rewrite emails or things of that nature?

00:25:07Well, you know, that 88%, 92% accuracy is fine.

00:25:13If we're looking at an underwriting scenario desk or an underwriting knowledge agent, like

00:25:18you need that to be at 99%, right?

00:25:21We, five, I'm sure Randall's over there.

00:25:23He's like, I'm not taking 8% wrong on my guidelines, right?

00:25:27From that perspective.

00:25:29And so I would say, does it need to be perfect?

00:25:32It really just comes down to what is the solution that you're trying to solve?

00:25:36But I would say to folks, you know, really start off with something, you know, that's

00:25:42internal and that may be a small step.

00:25:45So that could be something like I mentioned earlier, like a voice coach, that could be

00:25:50something like a knowledge agent, or it could be something, you know, I've even seen folks

00:25:56who are looking at like, I just, I have to have everybody checking the property address

00:26:01on the appraisal against USPS.

00:26:03And I just want to automate that whole piece and I want to let it go, right?

00:26:08And that's one that's, you know, again, it's fairly simple.

00:26:11So I definitely think it's a crawl, walk, run.

00:26:15And I think you just really need to look at, you know, what's your risk appetite on the

00:26:21problem that you're trying to solve for?

00:26:25Absolutely.

00:26:25Thank you so much.

00:26:26And go for further in.

00:26:29So absolutely.

00:26:30Yes.

00:26:30I mean, we see it can definitely augment underwriters with a different approach.

00:26:36It can be 80, 20.

00:26:38It can be 60, 40.

00:26:39It all depends on what we're trying to achieve with AI.

00:26:43And that's, it's happening in the market.

00:26:46So thank you so much for your...

00:26:49Sorry, Hadesha.

00:26:50There's one thing I just thought about that I do want to add is, is I think it's important

00:26:55to have conversations internally.

00:26:57Joy kind of touched on, you know, you're looking for what is the problem that I'm trying to

00:27:01solve, not just inject AI to inject AI.

00:27:05And oftentimes we find that there are things that folks are looking for that actually we

00:27:11probably could have built 10 years ago because it's not necessarily AI, but it's good.

00:27:16I think it's getting the industry talk, talking and thinking about it.

00:27:19But I would also really encourage folks have conversations with your underwriters, have

00:27:24conversations with your processors, your closers.

00:27:27Ask them, right?

00:27:28Like what gets in the way of your job on a day-to-day basis?

00:27:32Because you're also going to increase your adoption much higher if you're coming in with

00:27:37a tool that's helping them and that they understand and they see where that value is.

00:27:43Absolutely.

00:27:44Getting compliance and ops early on in the business, understanding what challenges are

00:27:57from an ops perspective, compliance perspective, definitely will go the long way.

00:28:04And there are certain things which technically you can take care in the processing side of

00:28:08the business before it reaches the upstream to underwriters, right?

00:28:12So it's not just underwriting.

00:28:14It's the downstream processes where we can look at.

00:28:18I've come across certain solutions where a POS technically sends back to a borrower asking

00:28:27for a letter of explanation.

00:28:28The processor is not doing that.

00:28:33It's the POS that is doing that, where it's reviewing the credit and sending auto email

00:28:40to the borrower saying that, hey, you know what?

00:28:42I only have four documents of the bank statement and four pages of the five pages that you claim

00:28:48that you have sent for the bank statement.

00:28:50So the majority of the work is done at the downstream processes before it reaches to the core underwriter

00:28:56to actually go or process to go and review and give an update or probably an approval.

00:29:04So absolutely, I think involving processors and other roles into a very early stage

00:29:11is definitely going to help and uplift productivity across underwriting.

00:29:17Thank you, Anthony, for your time.

00:29:20So, I mean, I would say, you know, you look at it as AI as a tool rather than AI as a driver,

00:29:29if I have to sum it up, right, where you look at certain processes to enhance this.

00:29:33And how are underwriters' role evolving?

00:29:39I think we have touched on this, but we can still go through some use cases,

00:29:44how it can help underwriters.

00:29:47Like if you go to the following slides, we have listed down key things that they can really focus on, right?

00:29:55That 80% that we spoke about, the twin brother strategy that we spoke about,

00:30:01to what percentage.

00:30:02You know, when lenders are shifting from being, you know, a checker, maybe, you know, from operations to innovation,

00:30:12what are the things that they need to look at and how it's going to help the other persona.

00:30:17So, document is one thing, right?

00:30:20Getting the right data out to underwriters, running those models for income,

00:30:26checking information on bank statement, if there was any heavy, you know, large deposit,

00:30:31you know, before the application was fine, to check, you know, the source of that income.

00:30:40Checking fraud, if all these things are taken care by AI,

00:30:44so what is left with underwriters is technically focused on being, you know,

00:30:50a strategic risk analyst than just going through all the details.

00:30:53So, definitely, it's going to give a shift to the underwriter.

00:30:57I don't know, probably technically an underwriter today, Randall,

00:31:00would be doing, what, 2.5 to 3 files, fresh files a day?

00:31:04Is that accurate?

00:31:06Somewhere close?

00:31:09Yeah, I'm sorry.

00:31:10You cut out on me.

00:31:11I couldn't hear you at that point.

00:31:13I think today an underwriter would probably do 2.5 to 3 fresh files a day.

00:31:19Is that correct?

00:31:19Yeah, absolutely.

00:31:20Yeah, and then depending on the space, right,

00:31:23if you introduce non-QM or you're introducing a portfolio product

00:31:28that has a little bit more strenuous guidelines,

00:31:32they could take a little bit longer.

00:31:33So, you could be looking even at 2, 2.5, you know,

00:31:37depending on the complexity of those loans.

00:31:42Absolutely.

00:31:43So, I think definitely with, you know, with AI,

00:31:47you know, it's going to give lift to their productivity.

00:31:50A, it's going to make them more,

00:31:52more, you know, predictive than being reactive

00:31:57because some of the work would be done at a fraud level

00:32:00or income calculation level or, you know,

00:32:02checking some of the data fields from an appraiser.

00:32:05You know, all that work would be done.

00:32:08So, they can focus more on being more proactive than reactive.

00:32:12It definitely will make them, I would say, a knowledge, you know,

00:32:17a hub or a knowledge worker than just going through a checklist

00:32:21and, you know, chasing the volumes, right?

00:32:25Making sure that there are zero buybacks that are coming

00:32:28and there are no errors that are coming in post-close, right?

00:32:32Usually, all the work which is done, you know, from a QC perspective,

00:32:38although we say it's probably pre-underwriting,

00:32:42but majority of the QC errors are caught in post-close.

00:32:45So, if there are, you know, AI QC built into the system

00:32:49that helps underwriters, you know, create those templates,

00:32:52understand, you know, from a document,

00:32:55from a credit worthiness perspective,

00:32:57it definitely, we feel it's going to give lift to the underwriter.

00:33:02So, the new role of an underwriter would be, you know,

00:33:05more towards the strategic side is what we feel.

00:33:10Any comments, Randall, from your side?

00:33:13Yeah, no, I completely agree with what you're saying.

00:33:15Totally agree.

00:33:19Perfect.

00:33:19And we're not replacing the underwriters here.

00:33:21I mean, in the overall approach that we've focused so far,

00:33:25there is no replacement, right?

00:33:27Because humans, irrespective of whether it is underwriting, processing,

00:33:32has to be there.

00:33:33That's the base of all the models.

00:33:36It's just going to enhance their productivity.

00:33:38It's going to augment their road shifting from doing all the manual tasks

00:33:42and leaving it up to the OCRs and the NLP of the world

00:33:47to do those kind of tasks.

00:33:51Yeah, I totally agree.

00:33:52And to sort of tag on to that, you know, we know in our world,

00:33:56one plus one equals two.

00:33:58But sometimes the file is not that black and white.

00:34:02Sometimes a borrower scenario cannot be interpreted by a guideline

00:34:08that's written in black and white.

00:34:10And that's what we always refer to affectionately as the gray areas.

00:34:14And our interpretation of the risk in those gray areas is really where I can see

00:34:21that human underwriters will have to get involved.

00:34:24As AI is introduced more predominantly into an underwriting space,

00:34:29we have to teach AI what happens in those gray areas and what determined that it was a gray area.

00:34:37What said that that borrower circumstance made it okay,

00:34:42even though something in black and white could have been interpreted as not okay.

00:34:46And so I think that's definitely where the human aspect has to be involved.

00:34:51And our knowledge, our skill set as empathetic humans that understand circumstances

00:34:59and have experienced our own personal circumstances in our lives

00:35:04have to train an artificial intelligence what that looks like

00:35:08and how that becomes a reality in our current world.

00:35:12True, absolutely.

00:35:17I mean, yes.

00:35:19And, you know, if the models can predict delinquency and default at a very early stage,

00:35:25that's a game changer again, not just for, you know, an underwriter,

00:35:30but for the organization as a whole, right?

00:35:33You're detecting things at a very early stage,

00:35:37which really helps the underwriter, you know, take that call quite early in the entire process.

00:35:43And the touch rate for an underwriter definitely improves with these AIs, right?

00:35:52It will never be like a three-touch or a four-touch fight.

00:35:55You know, we can look at maybe at the max one or two.

00:35:58And that's where the productivity lift of, you know, 2.503, depending on the loan type.

00:36:05Like we said, non-QM probably would be more manual, so it takes more time.

00:36:11But, you know, an underwriter may be looking at four files, five files.

00:36:16It's definitely going to give a lift because when the volume goes up,

00:36:19you don't have to fire people, you know, hire people and then fire when the volumes are low, right?

00:36:25These automations, that's why I said it can be overwhelming at times,

00:36:28but at the same time, if we use them at the right place,

00:36:32it will definitely give us lift when the volumes are high, right?

00:36:36And underwriter being, you know, the most expensive resource across origination,

00:36:43I think some lift will definitely help from the P&L perspective.

00:36:50I do want to add on to Randall's point.

00:36:53I think one of the most important things to remember about AI is it has no viewpoint, right?

00:36:58Like we talk about, someone talked about this on LinkedIn the other day

00:37:02where they were talking about how, you know, the current version of GPT

00:37:06can pass the, you know, CPA exam or whatever it is, like 99%, right?

00:37:13It's smart, it can do a lot, but it has no viewpoint.

00:37:17And so to Randall's point, you know,

00:37:19when you have those loans that have compensating factors,

00:37:22I remember trying to automate compensating factors.

00:37:25Randall will probably laugh at this.

00:37:26Several years back and we just threw our hands up and we were like,

00:37:30it's literally impossible because there's just so many scenarios

00:37:34and it requires that human intervention who has that viewpoint

00:37:39to look at all of these different things.

00:37:42Because the reality is, is, you know,

00:37:44we'd love to believe that our industry is black and white, but it's not.

00:37:47It's just a whole bunch of gray.

00:37:49And so I just don't see the underwriter ever being able to be replaced entirely

00:37:57because you're always going to need a human in the loop to handle those scenarios.

00:38:03I totally agree.

00:38:06And the early we involve underwriters in the automation process,

00:38:12the better for the organization as well, for the models to be successful.

00:38:18The way, you know, documents have a confidence score,

00:38:22the same way I think humans, when they work on new technology,

00:38:25want to get the confidence score, how good the system is.

00:38:29Am I getting the right result?

00:38:30Otherwise, I mean, we might deploy a tool, you know,

00:38:34assuming that it's going to give lift to the underwriter,

00:38:38but the underwriter is, you know, still doing everything manual,

00:38:41not trusting the system, right?

00:38:44Because for any system, for any deployment, it takes time.

00:38:47It takes time.

00:38:48It takes time to reach to that accuracy level.

00:38:51You probably start at 40, eventually go to 60, 80,

00:38:54and then maybe, you know, reach that level where an underwriter will say,

00:38:58you know what, this section, I know it will be accurate.

00:39:01I've seen it grow over the last three, four months,

00:39:03you know, the accuracy going up.

00:39:05So the more we involve, you know, underwriters or processors

00:39:10in the entire automation process of how things work,

00:39:13I feel, you know, and that's what we've seen

00:39:15in some of our consulting engagements, it gives better results.

00:39:19In fact, during the course of the, you know, deployment,

00:39:23there are a lot of, you know, changes that have been brought up.

00:39:27You know, not just from that persona,

00:39:30but different personas of how to make it better.

00:39:33Because we look at that particular function

00:39:37from an automation perspective,

00:39:39but when we involve different personas,

00:39:41we get to know that, oh, this is the challenge

00:39:44that happens at the upstream process

00:39:47when the file goes to a closer.

00:39:49So this is what the processor is facing an issue,

00:39:51and that's why he's not able to fix it,

00:39:53but just pass it on to the underwriter.

00:39:55And the underwriter is again going to set it back,

00:39:57saying that, you know, conditions on the file,

00:39:59I don't have this.

00:40:00So that helps a lot.

00:40:02So definitely it is going to, you know,

00:40:05help the underwriters,

00:40:06and the new role of the underwriter

00:40:07would be much different

00:40:09than what the traditional role has been so far.

00:40:12And I'd like to add to that, Hatish,

00:40:16is that we're going into this assuming

00:40:18that these individuals are willing to change.

00:40:21And what I have found in my experience

00:40:23is that there is an adoption rate component

00:40:26to be considered.

00:40:27And I see these customers,

00:40:29these companies who are like,

00:40:31let's go to the cloud.

00:40:32Let's go with this automation.

00:40:34Let's innovate this technology and this tech stack.

00:40:37But then when they are introduced,

00:40:40and a good example of that

00:40:41is going from a form-based software

00:40:44to a data-driven software.

00:40:47And they're so comfortable with,

00:40:48even in, gosh, I think it was 2015

00:40:51when lenders were asked to move

00:40:53from paper files to paperless.

00:40:56And so we were making that shift,

00:40:57and I'm still seeing customers

00:40:59who even today still prefer

00:41:02those handwritten files.

00:41:03And it's shocking to me

00:41:04because I'm like, that was 10 years ago.

00:41:06But that adoption rate is a real thing

00:41:09and has to be considered

00:41:10because once you introduce

00:41:11this new technology in this new way,

00:41:14no matter how efficient it may be,

00:41:16it's really hard to accept it,

00:41:19adopt it, and start the practice

00:41:21and to move forward with it.

00:41:23Another example is that we have

00:41:25automated conditions and fees

00:41:27and validation checks.

00:41:29And yet you might have a customer who says,

00:41:31but I like my Excel sheet

00:41:33that is my manual checklist.

00:41:35And so it's really hard for individuals,

00:41:39intelligent, really great at their job individuals,

00:41:42who can grasp or just let go

00:41:45of those comfort zones

00:41:47and where they have that control

00:41:49and they can be sure that that files,

00:41:51you know, crossing all the T's,

00:41:52dotting all the I's.

00:41:53And so that adoption rate

00:41:54is definitely something to consider

00:41:56and applies back to what Andrew said,

00:41:58the crawl, walk, run phase.

00:42:00We're still in that crawl phase

00:42:01from what I'm seeing

00:42:02with lenders and individuals.

00:42:04And it is hard to make a change.

00:42:06It is hard to accept these things,

00:42:08but we are making progress

00:42:10and we're going to get there,

00:42:11but I do still think

00:42:12we're in that crawl phase.

00:42:15Absolutely.

00:42:15That's a very valid point,

00:42:17I would say, right?

00:42:17Change management.

00:42:19I mean, for me,

00:42:20it's like a silent killer, right?

00:42:22For AI adoption.

00:42:25Yeah.

00:42:26Data being one,

00:42:28you know, having the clean data

00:42:29to run the models,

00:42:30but then the change part,

00:42:32enjoy that you test on,

00:42:34you know,

00:42:35that's why I think we see

00:42:36the adoption rate,

00:42:37it's less,

00:42:38a lot of the organizations

00:42:39fall in the more than 50% bracket

00:42:42who are in the trial phase,

00:42:44right?

00:42:44They're thinking of it,

00:42:46of deploying,

00:42:47some are still thinking

00:42:48whether we should use OCR,

00:42:50or ICR,

00:42:51follow the same paper heavy process.

00:42:55Right.

00:42:56So I think that makes

00:42:57a big difference, right?

00:42:59Adoption.

00:43:00And what we've seen,

00:43:01you know,

00:43:02in some of the consulting engagements

00:43:04is the success

00:43:08depends on,

00:43:10you know,

00:43:11more on the strategic part

00:43:12than the operational.

00:43:13The failure is very rarely

00:43:15on the operational side.

00:43:16It's more on the strategic side,

00:43:18right?

00:43:19Taking that decision

00:43:20that yes,

00:43:20we want to implement this

00:43:22in a particular way.

00:43:24And how do you want

00:43:24to implement it?

00:43:26Right?

00:43:26Rather than just,

00:43:27you know,

00:43:28staying neutral

00:43:30like what we do,

00:43:31we don't know

00:43:31and half of the organization

00:43:33wants to apply AI

00:43:35because like I said,

00:43:36it's overwhelming.

00:43:37You read about AI

00:43:38everywhere

00:43:39and then you see,

00:43:41you hear these big names

00:43:43who are using AI

00:43:44and, you know,

00:43:45having a cycle time

00:43:46of 10 days

00:43:47and 15 days,

00:43:48then as an organization

00:43:50you feel,

00:43:50you know,

00:43:51what is it

00:43:51that I'm not doing right?

00:43:52Why is it taking me

00:43:5335 or 40,

00:43:5550 days

00:43:56to close a loan?

00:43:57But I see,

00:43:58you know,

00:43:58the top lenders

00:43:59is closing it in 20 days

00:44:00and that's when

00:44:02the pressure starts

00:44:02building up.

00:44:03Right, Randall?

00:44:04Correct me for more.

00:44:06Absolutely.

00:44:07Well,

00:44:07and I think too,

00:44:09you know,

00:44:09speaking to Joy's point,

00:44:11a lot of lenders today

00:44:13were fighting

00:44:13for the same loan.

00:44:15You know,

00:44:15lenders are all

00:44:17going after the same loan

00:44:18for that qualified borrowers.

00:44:20Borrowers are being challenged

00:44:21in qualifying

00:44:22because DTIs are high

00:44:24as the rates have been high,

00:44:26creates,

00:44:27you know,

00:44:27challenges for qualifying.

00:44:30And then to that,

00:44:31so we're fighting

00:44:32over the same loans.

00:44:33Well,

00:44:33margins are low.

00:44:34So you have

00:44:35less profit

00:44:36happening within companies.

00:44:38So then they have

00:44:39cut headcount

00:44:39because volume is down.

00:44:41And then we introduce

00:44:43factors like AI

00:44:45that says,

00:44:46well,

00:44:46you could have

00:44:47less headcount

00:44:48because you could have

00:44:49the lift from AI.

00:44:51But now we pose

00:44:52the question of

00:44:53how do I implement that

00:44:54when I don't,

00:44:55when I'm just trying

00:44:56to get through the loans

00:44:57because we have less people.

00:44:58So I feel like

00:44:59there's probably

00:45:00as a company,

00:45:01there has to be

00:45:02a priority

00:45:02that is placed

00:45:04in AI

00:45:04for the lift

00:45:05that you will

00:45:06eventually get

00:45:07by having it.

00:45:08And probably

00:45:10why we are

00:45:11in that crawl phase

00:45:12being because

00:45:14there are no resources

00:45:15to throw at AI

00:45:17and get it implemented

00:45:19effectively

00:45:19in the company model.

00:45:22So that's

00:45:23definitely one of the challenges

00:45:24I could speak to today.

00:45:26We do not have

00:45:27a plethora

00:45:28of project managers

00:45:29on staff.

00:45:31So how we would implement

00:45:33and how we will implement

00:45:34AI in our space

00:45:36has to be factored

00:45:37into everybody's

00:45:38current day-to-day business

00:45:40and what they have

00:45:41to get finished

00:45:42in their business.

00:45:43So unless somebody amazing

00:45:44like Sutherland

00:45:45comes along

00:45:46that creates

00:45:47an avenue for AI

00:45:49and helps us

00:45:50with that lift,

00:45:51then certainly,

00:45:52you know,

00:45:53it becomes a challenge

00:45:54and probably why

00:45:55I feel like

00:45:55we are more

00:45:56in a crawl phase

00:45:57to how do I get it going?

00:45:59How do I make this start

00:46:00and happen?

00:46:04Absolutely.

00:46:05Andrew,

00:46:05do you hear that?

00:46:07I mean,

00:46:07when you're discussing,

00:46:08you know,

00:46:09AI in these conferences

00:46:11or with your customers,

00:46:12do you see

00:46:13the same challenges?

00:46:15Yeah,

00:46:15100%.

00:46:16I mean,

00:46:16this is the first

00:46:17technological revolution

00:46:18where we anticipate

00:46:19everyone to suddenly

00:46:21be a CTO.

00:46:22you know,

00:46:23I spoke at the New England

00:46:25Mortgage Bankers Conference

00:46:26two weeks ago

00:46:27and there's,

00:46:29I don't know,

00:46:29maybe 120,

00:46:30130 people in the room

00:46:31and I asked,

00:46:32how many of you

00:46:33are from the technology

00:46:34department in your company?

00:46:36Two people.

00:46:37Two people raised their hands.

00:46:38That means that

00:46:39every other single person

00:46:40in the room

00:46:40was in sales

00:46:41or operations

00:46:42and that really

00:46:44should say something,

00:46:45right?

00:46:45Because like,

00:46:46you know,

00:46:46if 10 years ago

00:46:47we were having a talk

00:46:49on, you know,

00:46:49cloud-based technology,

00:46:51you know,

00:46:51there wouldn't be

00:46:52any ops or sales folks,

00:46:54you know,

00:46:55that are showing up

00:46:55into the room.

00:46:57So,

00:46:57I think it is critical

00:47:00right now.

00:47:03What I recommend

00:47:04to a lot of folks

00:47:05is when you're looking

00:47:06at AI vendors,

00:47:08look at AI vendors

00:47:09who have a track record

00:47:10of being in the mortgage industry.

00:47:13People who speak

00:47:14the language,

00:47:15who understand

00:47:16the processes

00:47:17and I don't mean to,

00:47:18you know,

00:47:19there are a lot of

00:47:19really brilliant people

00:47:20who are coming out

00:47:21of San Francisco

00:47:22and New York

00:47:23with some great ideas

00:47:25but the struggle is

00:47:26and it's just the reality

00:47:27is they don't speak

00:47:28the language necessarily

00:47:30and so it does take

00:47:31a little bit more lift

00:47:33from someone like Randall

00:47:35in trying to get

00:47:36that implemented

00:47:37on that side.

00:47:40I do want to add

00:47:41one thing around adoption

00:47:43and this is just

00:47:44our philosophy

00:47:45and we didn't come up

00:47:47with it.

00:47:47I stole it

00:47:48from Simon Sinek

00:47:49so he gets

00:47:50all the credit here

00:47:51but what we really

00:47:52try to do

00:47:53with our organization

00:47:54or with our clients

00:47:55or recommend

00:47:56is let's find

00:47:57a few early adopters

00:47:58within your organization

00:48:00and let's do

00:48:01a proof of concept,

00:48:03right?

00:48:03We'll do a free

00:48:04proof of concept

00:48:05with those individuals

00:48:07to get them

00:48:08really excited,

00:48:09right?

00:48:09Because those

00:48:09are the people

00:48:10who are again

00:48:11are early adopters

00:48:12are going to get excited

00:48:13about new technology

00:48:14and then once

00:48:16they get it implemented

00:48:17and they start using it

00:48:19then the sort of middle

00:48:21the fence sitters

00:48:22if you will

00:48:23all of a sudden

00:48:24start watching

00:48:25the early adopters

00:48:26and they're like

00:48:26ooh, I want that

00:48:27that looks really cool

00:48:28and the next thing

00:48:29you know

00:48:29you've got 60 to 70%

00:48:31of your organization

00:48:32using this new tool

00:48:35and inevitably

00:48:36what happens

00:48:37is the people

00:48:39who are

00:48:39the more difficult

00:48:41to adopt

00:48:42suddenly get jealous

00:48:43of the other

00:48:4470%, right?

00:48:46And they're like

00:48:46oh, I want that toy

00:48:47because everybody else

00:48:48has that toy

00:48:49so I do think

00:48:50that's another

00:48:50kind of crawl, walk, run

00:48:52when we're talking

00:48:53about the people

00:48:54within your organization

00:48:55to really help

00:48:57with that adoption

00:48:57but also

00:48:59it helps for you

00:49:00to prove out

00:49:01the ROI

00:49:02right?

00:49:03When you're

00:49:03when you're looking down

00:49:04and seeing

00:49:05each

00:49:06each group

00:49:08adopting it

00:49:09through the process

00:49:10thanks

00:49:14thanks Andrew

00:49:14for that insight

00:49:15great

00:49:17I think

00:49:18good days

00:49:20for underwriters

00:49:21if we move on

00:49:23to our next

00:49:24question

00:49:26that we want

00:49:26to touch on

00:49:27which is

00:49:28can AI

00:49:29make compliance

00:49:30effortless?

00:49:32Right

00:49:32we purposely

00:49:33put this

00:49:33question

00:49:35because there's

00:49:36so much

00:49:36that's been written

00:49:37now we understand

00:49:39that AI

00:49:40is rapidly

00:49:40reshaping

00:49:41the mortgage

00:49:42landscape

00:49:42and

00:49:43effortless

00:49:45I think

00:49:45is an aspiration

00:49:47is what

00:49:49you know

00:49:50we feel

00:49:50it's an aspiration

00:49:51but

00:49:52webinars like this

00:49:54I think

00:49:54it's going to help

00:49:55us understand

00:49:57and you know

00:49:58those who are listening

00:49:59understand that

00:50:00how close

00:50:01are we getting

00:50:01to you know

00:50:03to the effortless

00:50:04part of compliance

00:50:05does it really

00:50:05help compliance

00:50:07in a way

00:50:07it does

00:50:08we have to wait

00:50:10and watch

00:50:11and see

00:50:11you know

00:50:12whether it is

00:50:12effortless

00:50:13or you know

00:50:13what shift

00:50:14that it helps

00:50:15how much

00:50:16of an effort

00:50:18will it reduce

00:50:18you know

00:50:19dramatically

00:50:20will help

00:50:21institutions

00:50:21stay ahead

00:50:22of regulations

00:50:23as we move

00:50:26to some

00:50:28data points

00:50:28we've seen

00:50:30some studies

00:50:31where

00:50:32organizations

00:50:33are adopting

00:50:34AI

00:50:35in compliance

00:50:36again

00:50:37you know

00:50:39going back

00:50:39to

00:50:40to the point

00:50:41that

00:50:42taking a step

00:50:42back

00:50:43understanding

00:50:44what is

00:50:45required

00:50:46who do you

00:50:47involve

00:50:48from the

00:50:48initial phase

00:50:49what governance

00:50:51model do you

00:50:52do you set

00:50:53to review this

00:50:54because you again

00:50:55don't want

00:50:56to be

00:50:56you know

00:50:58have a situation

00:50:59like for example

00:51:00if a borrower

00:51:02is denied

00:51:02a loan

00:51:03and

00:51:04an AI

00:51:05is

00:51:05you know

00:51:06sending

00:51:06not sending

00:51:08an explanation

00:51:08to the borrower

00:51:10why it was

00:51:11denied

00:51:11would probably

00:51:12put the lender

00:51:13into

00:51:13you know

00:51:14into a fix

00:51:15where you're

00:51:16supposed to

00:51:17give an

00:51:17explanation

00:51:18on why

00:51:18so you

00:51:18don't want

00:51:19to leave

00:51:19certain things

00:51:20to AI

00:51:21from a

00:51:21compliance

00:51:22perspective

00:51:22you still

00:51:23want humans

00:51:24to focus

00:51:25on those

00:51:25critical parts

00:51:26but yes

00:51:27AI can

00:51:28definitely

00:51:29help

00:51:30in

00:51:30you know

00:51:31getting

00:51:31the red

00:51:32flags

00:51:33on fraud

00:51:33when it

00:51:34comes to

00:51:34document

00:51:35identity

00:51:36data

00:51:36and help

00:51:38the compliance

00:51:39department

00:51:40to fix

00:51:41certain things

00:51:42and being

00:51:43more proactive

00:51:44again than

00:51:44being you know

00:51:45reactive

00:51:46after the loan

00:51:46is closed

00:51:47and then trying

00:51:47to do

00:51:48a due

00:51:49diligence

00:51:49because

00:51:49in mortgages

00:51:51what I've

00:51:52seen

00:51:52you know

00:51:53is after

00:51:54every step

00:51:55there's a QC

00:51:56right

00:51:58when the

00:51:59processing

00:51:59is done

00:51:59there's a

00:52:00QC

00:52:00before the

00:52:01file goes

00:52:01to underwriting

00:52:02there's a

00:52:02QC

00:52:02after the

00:52:03loan is

00:52:04closed

00:52:04we do

00:52:04post-close

00:52:06QC

00:52:06once the

00:52:06loan is

00:52:07sold out

00:52:07you know

00:52:08correspondent

00:52:09there's

00:52:09another

00:52:09QC

00:52:10then you

00:52:10have those

00:52:11third-party

00:52:11review

00:52:12sites

00:52:12so the

00:52:13QC

00:52:13which is

00:52:13happening

00:52:14again

00:52:14to make

00:52:15sure

00:52:15that you're

00:52:16compliant

00:52:16so

00:52:18definitely

00:52:19from that

00:52:20perspective

00:52:20AI can

00:52:21provide

00:52:22that lift

00:52:23help

00:52:24you

00:52:25you know

00:52:25being the

00:52:26green zone

00:52:27with the

00:52:27regulators

00:52:28make sure

00:52:28that you're

00:52:29not

00:52:30you know

00:52:31an outlier

00:52:33when it

00:52:33comes to

00:52:33trade

00:52:34and

00:52:34other

00:52:36regulations

00:52:37that you

00:52:38have to

00:52:38follow

00:52:39and

00:52:40I would

00:52:42love to

00:52:43hear your

00:52:44point of

00:52:44view

00:52:45when it

00:52:45comes to

00:52:46compliance

00:52:46what are

00:52:48your thoughts

00:52:48I mean

00:52:48and this

00:52:49not necessarily

00:52:50in origination

00:52:51it can be

00:52:52across the

00:52:52life cycle

00:52:53how do

00:52:54you see

00:52:55AI

00:52:55helping

00:52:56compliance

00:52:57yeah

00:52:58so it's

00:52:59definitely

00:52:59not

00:52:59effortless

00:53:00and AI

00:53:01is definitely

00:53:02not going

00:53:03to replace

00:53:03compliance

00:53:04or attorneys

00:53:06as much

00:53:07as we may

00:53:07wish

00:53:08on that

00:53:09one

00:53:10I think

00:53:11there's a

00:53:11lot

00:53:11but I do

00:53:12think

00:53:12there's a lot

00:53:12of applications

00:53:13here

00:53:13I do

00:53:14think

00:53:15you know

00:53:15one of

00:53:16the things

00:53:16that even

00:53:17I've leveraged

00:53:18it for

00:53:18on the

00:53:19compliance

00:53:19is to

00:53:21better

00:53:21explain

00:53:22regs

00:53:23to

00:53:23individuals

00:53:24within

00:53:24the

00:53:25organization

00:53:25right

00:53:26these

00:53:26things

00:53:26can be

00:53:26incredibly

00:53:27difficult

00:53:27to read

00:53:28the lawyers

00:53:30do that

00:53:30on purpose

00:53:31so then

00:53:31you have

00:53:31to hire

00:53:31them to

00:53:32read them

00:53:32back to

00:53:33you

00:53:33and explain

00:53:34it

00:53:34so that

00:53:35is

00:53:35one way

00:53:36that I

00:53:36think

00:53:36we're

00:53:37seeing

00:53:37it

00:53:37when it's

00:53:38leveraged

00:53:38on the

00:53:39QC

00:53:40point

00:53:40and I'm

00:53:41actually

00:53:41going to

00:53:42get away

00:53:42from AI

00:53:42on this

00:53:43one

00:53:43for a

00:53:43moment

00:53:44coming

00:53:44from

00:53:45the

00:53:45third

00:53:46party

00:53:46review

00:53:46world

00:53:47and mortgage

00:53:47backed

00:53:48securities

00:53:48is

00:53:48eventually

00:53:49we need

00:53:50to move

00:53:50to some

00:53:51sort

00:53:51of a

00:53:51blockchain

00:53:52when it

00:53:53comes to

00:53:54these

00:53:54QC

00:53:54processes

00:53:55because

00:53:56you're

00:53:57100%

00:53:57correct

00:53:58these

00:53:59loans

00:53:59are

00:53:59just

00:54:00getting

00:54:00QC

00:54:00over

00:54:01and over

00:54:02and over

00:54:02again

00:54:03and it

00:54:04just

00:54:04I think

00:54:05that's

00:54:06a big

00:54:06opportunity

00:54:08for someone

00:54:09to really

00:54:10make a

00:54:11splash

00:54:11in the

00:54:13industry

00:54:13if they

00:54:13can solve

00:54:14a problem

00:54:15because

00:54:16it's

00:54:16extremely

00:54:17costly

00:54:17right

00:54:18and the

00:54:19question

00:54:20is

00:54:20again

00:54:22coming

00:54:22from the

00:54:23secondary

00:54:23world

00:54:24that I

00:54:24always

00:54:24have

00:54:24is

00:54:25why

00:54:25do

00:54:25we

00:54:26wait

00:54:26so

00:54:26long

00:54:27to find

00:54:29out

00:54:29if a

00:54:29loan

00:54:30is

00:54:30securitizable

00:54:31we're

00:54:32waiting

00:54:32until

00:54:33three days

00:54:34or a

00:54:34week

00:54:35after the

00:54:35loan

00:54:36actually

00:54:36closes

00:54:36and then

00:54:38we're

00:54:38shipping

00:54:38it off

00:54:38to a

00:54:39due

00:54:39diligence

00:54:39firm

00:54:40to now

00:54:40do a

00:54:41complete

00:54:41100%

00:54:42re-underwrite

00:54:43to then

00:54:44send it

00:54:44off

00:54:45then if

00:54:47the

00:54:47MSR

00:54:48gets

00:54:48sold

00:54:48off

00:54:48of

00:54:49the

00:54:49loan

00:54:49now

00:54:49it

00:54:49gets

00:54:50reviewed

00:54:50again

00:54:50and

00:54:51then

00:54:51if

00:54:51there's

00:54:51an

00:54:51aged

00:54:51asset

00:54:52trade

00:54:52it gets

00:54:53done

00:54:54again

00:54:54if

00:54:55the

00:54:55securitization

00:54:56gets

00:54:56re-levered

00:54:57we gotta

00:54:57look at

00:54:58the file

00:54:58again

00:54:58and it

00:54:59just

00:54:59doesn't

00:55:00make

00:55:00any

00:55:01sense

00:55:01and I

00:55:01think

00:55:01moving

00:55:02again

00:55:02sorry

00:55:03to get

00:55:03off

00:55:03the

00:55:03AI

00:55:04component

00:55:04here

00:55:05but I

00:55:05think

00:55:05moving

00:55:05to

00:55:06some

00:55:06sort

00:55:06of

00:55:06blockchain

00:55:07which

00:55:08some

00:55:08folks

00:55:08are

00:55:08starting

00:55:09to

00:55:09do

00:55:09we

00:55:09are

00:55:15that

00:55:15in

00:55:15the

00:55:15securitization

00:55:16space

00:55:16I

00:55:17think

00:55:18that's

00:55:18going

00:55:18to

00:55:18help

00:55:19to

00:55:19significantly

00:55:20reduce

00:55:20that

00:55:21now

00:55:21all

00:55:22of

00:55:22that

00:55:22being

00:55:22said

00:55:22I

00:55:23do

00:55:23think

00:55:23from

00:55:24a

00:55:24compliance

00:55:24perspective

00:55:25other

00:55:26applications

00:55:26of

00:55:27artificial

00:55:27intelligence

00:55:28will

00:55:28definitely

00:55:28start

00:55:30to

00:55:30pop

00:55:31up

00:55:31where

00:55:32again

00:55:33we're

00:55:33able

00:55:33to run

00:55:33maybe

00:55:34some

00:55:34of

00:55:34the

00:55:34fraud

00:55:34checks

00:55:35that

00:55:35traditionally

00:55:36have

00:55:37been

00:55:38very

00:55:38manual

00:55:39and

00:55:40as

00:55:40we're

00:55:40collecting

00:55:40all

00:55:41that

00:55:41data

00:55:41we

00:55:42can

00:55:42leverage

00:55:42AI

00:55:43to

00:55:43do

00:55:43the

00:55:44data

00:55:44comparisons

00:55:45to

00:55:46run

00:55:46for

00:55:47those

00:55:47quality

00:55:47checks

00:55:48right

00:55:48because

00:55:48honestly

00:55:49a lot

00:55:49of

00:55:49times

00:55:50that's

00:55:51what

00:55:51it

00:55:51ends

00:55:51up

00:55:52being

00:55:52it's

00:55:53just

00:55:53getting

00:55:53to

00:55:54that

00:55:54step

00:55:54of

00:55:54I

00:55:55think

00:55:55the

00:55:55unmutable

00:55:56data

00:55:56thank

00:55:59you

00:55:59thank

00:56:00you

00:56:00for

00:56:00the

00:56:00insight

00:56:01Andrew

00:56:01would

00:56:02be

00:56:02interesting

00:56:03to

00:56:03know

00:56:03Andrew

00:56:03from

00:56:04a

00:56:04bank's

00:56:04perspective

00:56:05the

00:56:06way

00:56:06you

00:56:07look

00:56:07at

00:56:07compliance

00:56:08do

00:56:08you

00:56:08think

00:56:08it's

00:56:09more

00:56:09manual

00:56:09there

00:56:10are

00:56:10some

00:56:11not

00:56:12I

00:56:12mean

00:56:12not

00:56:12necessarily

00:56:13AI

00:56:13it

00:56:14can

00:56:14be

00:56:15other

00:56:15automations

00:56:15as well

00:56:16that help

00:56:18I mean

00:56:18how do

00:56:18you look

00:56:19at this

00:56:19yeah

00:56:20I

00:56:21completely

00:56:21agree

00:56:22with what

00:56:22Andrew

00:56:22said

00:56:23obviously

00:56:24from a

00:56:24banking

00:56:24perspective

00:56:25we

00:56:26pay

00:56:26right

00:56:27so we

00:56:27have

00:56:28underwriters

00:56:28that we

00:56:29pay

00:56:29then we

00:56:30have

00:56:31QC

00:56:31that we

00:56:32pay for

00:56:32pre-funding

00:56:33QC

00:56:34then we

00:56:35pay for

00:56:35post-closing

00:56:36QC

00:56:37and it

00:56:37could have

00:56:37been

00:56:38in the

00:56:39random

00:56:39selection

00:56:40of files

00:56:41that's

00:56:41required

00:56:42by the

00:56:42GSEs

00:56:43it could

00:56:43be that

00:56:44the loan

00:56:45that was

00:56:45PFQC

00:56:46that I've

00:56:46already

00:56:46paid for

00:56:47is now

00:56:48pre-funding

00:56:49QC

00:56:50that I'm

00:56:50paying for

00:56:51again

00:56:51and then

00:56:53in that

00:56:53when I

00:56:54go to

00:56:54sell

00:56:54my

00:56:55loans

00:56:55an

00:56:56aggregator

00:56:57is going

00:56:57to

00:56:57re-underwrite

00:56:58it

00:56:58and QC

00:56:59the loan

00:56:59and then

00:57:00if I

00:57:00sell it

00:57:00secondary

00:57:01Lord

00:57:02help me

00:57:03that the

00:57:03GSEs

00:57:04are now

00:57:04auditing

00:57:05the file

00:57:05and they

00:57:06are using

00:57:06AI technology

00:57:07that we

00:57:08did not

00:57:08have

00:57:08available

00:57:09to us

00:57:10and so

00:57:10now that

00:57:11QC

00:57:11looks

00:57:12completely

00:57:12different

00:57:12so to

00:57:13Andrew's

00:57:14point

00:57:14from a

00:57:15banking

00:57:16perspective

00:57:16we pay

00:57:17a lot

00:57:18of money

00:57:18and diligence

00:57:19a lot

00:57:20of money

00:57:20and diligence

00:57:21and so

00:57:21to have

00:57:23an industry

00:57:25let's just

00:57:25say as a

00:57:26whole

00:57:27that gets

00:57:28our hands

00:57:29around a

00:57:29technology

00:57:30that can

00:57:30be trusted

00:57:31and affirms

00:57:33the information

00:57:34in a quicker

00:57:35period of time

00:57:36could we

00:57:36then

00:57:37scale back

00:57:38on some

00:57:39of the

00:57:40other areas

00:57:41of QC

00:57:41that's

00:57:42happening

00:57:42what happens

00:57:43when a

00:57:43diligence

00:57:44firm

00:57:44is

00:57:46known

00:57:47well

00:57:48and has

00:57:48a great

00:57:49reputation

00:57:49across the

00:57:50industry

00:57:50and they

00:57:52can do

00:57:52a review

00:57:52of my

00:57:53loans

00:57:53and now

00:57:54I sell

00:57:54that loan

00:57:55to any

00:57:55investor

00:57:55on a

00:57:56reliance

00:57:56letter

00:57:56how could

00:57:57AI not

00:57:58work in

00:57:59the same

00:57:59you know

00:58:00could AI

00:58:01work in

00:58:01the same

00:58:02regard

00:58:02that would

00:58:04cut cost

00:58:05for us

00:58:05as a

00:58:05company

00:58:06as a

00:58:06bank

00:58:06thanks

00:58:12thanks

00:58:12for

00:58:12the

00:58:12inside

00:58:13yes

00:58:14so we

00:58:15we all

00:58:15agree

00:58:16and

00:58:16absolutely

00:58:16it's not

00:58:17going to

00:58:17make it

00:58:17effortless

00:58:18but definitely

00:58:19if we

00:58:19provide

00:58:19some

00:58:20some

00:58:20shift

00:58:21from

00:58:22you know

00:58:23process

00:58:23efficiency

00:58:24perspective

00:58:24make life

00:58:25really easy

00:58:26with regards

00:58:27to catching

00:58:28fraud a little

00:58:29early

00:58:29getting the

00:58:30right data

00:58:31clean data

00:58:32to be submitted

00:58:33to the

00:58:34regulators

00:58:34on time

00:58:35so I

00:58:36think we've

00:58:36covered

00:58:37the

00:58:38applicability

00:58:39of AI

00:58:40the use

00:58:42cases

00:58:43on

00:58:43how it

00:58:45can be

00:58:46deployed

00:58:46across

00:58:47origination

00:58:48we've

00:58:48discussed

00:58:49about the

00:58:50different

00:58:50AI

00:58:50maturity

00:58:51levels

00:58:52on what

00:58:53types

00:58:53of AI

00:58:53can be

00:58:54used

00:58:54from

00:58:55machine

00:58:56learning

00:58:56to

00:58:57gen

00:58:58AI

00:58:59in which

00:59:00areas

00:59:00maybe

00:59:01chatbot

00:59:01into

00:59:02verification

00:59:02and others

00:59:03how it's

00:59:05going to

00:59:05impact

00:59:06underwriters

00:59:06and their

00:59:07role

00:59:07how the

00:59:08new world

00:59:08looks like

00:59:09and does

00:59:10it have

00:59:10any meat

00:59:11in the

00:59:11compliance

00:59:12part of

00:59:12the section

00:59:13which is

00:59:13critical

00:59:14now I

00:59:16think it's

00:59:16more on

00:59:17how do

00:59:17we

00:59:17deploy that

00:59:20and I

00:59:20think we

00:59:21all have

00:59:21spoken about

00:59:22the

00:59:23crawl

00:59:23walk

00:59:24and run

00:59:25phase

00:59:26and

00:59:26technically

00:59:26that's

00:59:28the

00:59:28framework

00:59:29that

00:59:30Sutherland

00:59:30uses

00:59:32be it

00:59:33for AI

00:59:34be it

00:59:34for any

00:59:35technology

00:59:35that's

00:59:37the

00:59:37framework

00:59:38that we

00:59:38start

00:59:39with

00:59:39that we

00:59:39want to

00:59:40first

00:59:40understand

00:59:42the use

00:59:42cases

00:59:43pick up

00:59:44low end

00:59:46task

00:59:46where we

00:59:47see ROI

00:59:48we get

00:59:50the compliance

00:59:50the ops

00:59:51all the

00:59:52teams

00:59:52together

00:59:52from

00:59:53the beginning

00:59:55to make

00:59:56sure everyone

00:59:56is aligned

00:59:57to that

00:59:58you pick

00:59:58up maybe

00:59:58I would say

00:59:59income

00:59:59verification

01:00:00kind of

01:00:00a process

01:00:01or a

01:00:01fraud

01:00:01detection

01:00:02kind of

01:00:02a process

01:00:03pilot it

01:00:04test it

01:00:05see the

01:00:06success of

01:00:07it

01:00:07you might

01:00:08fail but

01:00:08again

01:00:09that is

01:00:10needed

01:00:10in the

01:00:11crawl

01:00:11phase

01:00:11so you

01:00:12pick up

01:00:12low end

01:00:13processes

01:00:13work on

01:00:14that

01:00:14maybe a

01:00:15chatbot

01:00:16to start

01:00:17with if

01:00:17it's not

01:00:18there

01:00:19and then

01:00:20see

01:00:20how you

01:00:22can implement

01:00:22that when

01:00:23you go to

01:00:23the walk

01:00:24phase

01:00:24you probably

01:00:25give more

01:00:25authority to

01:00:26the AI

01:00:26you can

01:00:27add

01:00:27Gen AI

01:00:28to the

01:00:28chatbot

01:00:29that can

01:00:30help the

01:00:31borrowers

01:00:31upload documents

01:00:32real time

01:00:33detect if

01:00:34it's a blank

01:00:35page

01:00:35it's a

01:00:35utility page

01:00:36is it

01:00:37really a

01:00:37pay stop

01:00:38that's the

01:00:39next level

01:00:39of walk

01:00:40field

01:00:40and then

01:00:41figure out

01:00:41if you

01:00:41really want

01:00:42to go

01:00:42in the

01:00:43in the

01:00:43agentic

01:00:44way

01:00:44right

01:00:45have you

01:00:46been

01:00:46confident

01:00:47enough

01:00:47to use

01:00:48use cases

01:00:49that are

01:00:49giving you

01:00:50ROI

01:00:50right

01:00:51and then

01:00:52move on

01:00:53to processes

01:00:53that will

01:00:54benefit you

01:00:54so definitely

01:00:56you know

01:00:57that's the

01:00:58framework

01:00:58that we

01:00:58follow

01:00:59and I

01:01:00think that's

01:01:00what everyone

01:01:01has resonated

01:01:02on the call

01:01:02that crawl

01:01:04walk and run

01:01:05is the

01:01:05way forward

01:01:06and the

01:01:09industry right

01:01:09now is in

01:01:10the crawl

01:01:10phase is

01:01:11what my

01:01:13perspective

01:01:13would be

01:01:14but I

01:01:15would like

01:01:16Andrew to

01:01:17second that

01:01:19maybe add

01:01:20your points

01:01:20on where

01:01:21do you

01:01:21see the

01:01:23industry

01:01:23when it

01:01:24comes to

01:01:24AI

01:01:24from a

01:01:27call phase

01:01:27from a

01:01:28walk phase

01:01:28because when

01:01:29we read

01:01:31reports

01:01:32we get

01:01:34to know

01:01:35that oh

01:01:35my god

01:01:35there are

01:01:36certain

01:01:36lenders who

01:01:37are in

01:01:38the walk

01:01:38phase or

01:01:39in the

01:01:39run phase

01:01:39but are

01:01:40they actually

01:01:40in the

01:01:40run phase

01:01:41and if

01:01:42they are

01:01:42then why

01:01:43is it not

01:01:44being replicated

01:01:45by other

01:01:46lenders

01:01:46so where

01:01:47exactly from

01:01:48your

01:01:48conversation

01:01:49where

01:01:50exactly do

01:01:50you feel

01:01:51the industry

01:01:52is and

01:01:53how fast

01:01:54is it

01:01:55progressing

01:01:55with AI

01:01:56so it's

01:01:59definitely

01:01:59in the

01:01:59crawl phase

01:02:00and if

01:02:01there's

01:02:02nothing else

01:02:02that people

01:02:04take away

01:02:04from this

01:02:04webinar

01:02:05you are

01:02:05not behind

01:02:06first of

01:02:11all AI

01:02:11is not

01:02:12magical

01:02:13fairy dust

01:02:14there is

01:02:16no shortage

01:02:17of headlines

01:02:18things that

01:02:20I question

01:02:21especially

01:02:22being deep

01:02:23in it

01:02:23every day

01:02:24you know

01:02:25it's hard

01:02:26right

01:02:27when you

01:02:27look at

01:02:28you know

01:02:28the people

01:02:29who are

01:02:29doing

01:02:30you know

01:02:31things in

01:02:31the AI

01:02:32space

01:02:33from technology

01:02:34partners

01:02:34they're very

01:02:35hyper

01:02:35all of us

01:02:36are very

01:02:37hyper

01:02:37focused

01:02:37on a

01:02:38niche

01:02:38right

01:02:39and most

01:02:40of us

01:02:40have been

01:02:40around

01:02:40for two

01:02:41three years

01:02:42focused

01:02:43on that

01:02:44niche

01:02:44so

01:02:45be careful

01:02:48what you

01:02:48hear from

01:02:49particularly

01:02:50publicly traded

01:02:51companies

01:02:51who have

01:02:53a fiduciary

01:02:53responsibility

01:02:54to their

01:02:55shareholders

01:02:55to make

01:02:57it sound

01:02:58really great

01:02:58what they're

01:02:59doing

01:02:59what I

01:03:02have found

01:03:02from people

01:03:03that I'm

01:03:03talking to

01:03:03I talked

01:03:04to a top

01:03:0420 lender

01:03:06three weeks

01:03:07ago and I

01:03:08was like hey

01:03:08what are you

01:03:08guys doing

01:03:09with AI

01:03:09because I

01:03:10really like

01:03:11to hear

01:03:11where people

01:03:12are at

01:03:12and they're

01:03:12like we're

01:03:13looking at

01:03:14voice coaching

01:03:14and I was

01:03:16like what

01:03:16what else

01:03:17what are you

01:03:17doing

01:03:18we're looking

01:03:19at voice

01:03:19coaching

01:03:19like this

01:03:20is a top

01:03:2120 lender

01:03:21who's done

01:03:22literally

01:03:22nothing

01:03:23they have

01:03:23not installed

01:03:24anything in

01:03:25their operations

01:03:26they decided

01:03:27that AI

01:03:27voice coaching

01:03:28is where

01:03:28they want

01:03:29to start

01:03:29and so

01:03:30that's where

01:03:31they're at

01:03:32now I'm

01:03:32not saying

01:03:32that's where

01:03:33everybody is

01:03:34at but

01:03:34that should

01:03:35you know

01:03:35just give

01:03:36everybody

01:03:36you know

01:03:37an idea

01:03:37where the

01:03:38temperature

01:03:38sort of

01:03:39is from

01:03:41what I'm

01:03:41seeing

01:03:42is it's

01:03:43you know

01:03:43we're solving

01:03:44for very

01:03:45small problems

01:03:46and again

01:03:46it depends

01:03:47on you know

01:03:48the organization

01:03:48right like

01:03:49do you have

01:03:50the bandwidth

01:03:50within your

01:03:51developer group

01:03:52do you not

01:03:54you know

01:03:54are you building

01:03:55some internal

01:03:56things but

01:03:56from what

01:03:58I've seen

01:03:58everybody's

01:03:59crawling

01:03:59or they're

01:04:02thinking about

01:04:02crawling

01:04:03so don't

01:04:05don't feel

01:04:06like I

01:04:06always say

01:04:06like if you

01:04:07feel like

01:04:07you're behind

01:04:08you're probably

01:04:09exactly where

01:04:10you should

01:04:10be right

01:04:12now and

01:04:13I would

01:04:13just say

01:04:14you know

01:04:14start having

01:04:15conversations

01:04:15with people

01:04:17would be my

01:04:18other

01:04:18recommendation

01:04:19you know

01:04:20for anyone

01:04:21watching this

01:04:22you know

01:04:22reach out

01:04:23to you know

01:04:24somebody like

01:04:24joy or

01:04:25hitesh or

01:04:26myself or

01:04:27you know

01:04:27another technology

01:04:28partner or

01:04:29consultant that

01:04:30you know

01:04:30and just start

01:04:31to have some

01:04:32conversations

01:04:32and see you

01:04:34know what are

01:04:35different people

01:04:35doing and

01:04:36and what

01:04:37resonates for

01:04:38you as

01:04:39as a good

01:04:39place to

01:04:40start in

01:04:40your

01:04:40organization

01:04:41absolutely

01:04:45thank you

01:04:46Andrew for

01:04:47that point

01:04:48of view

01:04:48Randall Joy

01:04:49any comments

01:04:50from your

01:04:51side on

01:04:52the approach

01:04:52I would

01:04:53like to

01:04:54first off

01:04:55say thank

01:04:56you Andrew

01:04:56because I

01:04:56feel like our

01:04:57industry really

01:04:57needs to hear

01:04:58exactly what

01:04:59you just said

01:05:00it is a

01:05:01comfort to

01:05:01know I'm

01:05:02not behind

01:05:03I'm where

01:05:04I need to

01:05:04be and I

01:05:05think that as

01:05:06they start

01:05:06they meaning

01:05:07anyone in

01:05:08this industry

01:05:09who really

01:05:09wants to get

01:05:10to that level

01:05:11and create

01:05:12that space

01:05:12for efficiency

01:05:13you do need

01:05:16to consider

01:05:16am I using

01:05:17a flexible

01:05:18modernized

01:05:19mortgage tech

01:05:20stack and

01:05:21in that thought

01:05:22process you

01:05:23also have to

01:05:24consider the

01:05:24compliance that

01:05:25we've talked

01:05:26about the

01:05:26underwriting

01:05:27efficiencies

01:05:28even the

01:05:28processing

01:05:29efficiencies

01:05:29having an

01:05:31LOS that

01:05:32comes with

01:05:33a POS

01:05:34improves the

01:05:34data integrity

01:05:35and security

01:05:37of that

01:05:37information so

01:05:38it's really

01:05:39important to

01:05:40be thoughtful

01:05:41of taking

01:05:42those steps

01:05:43and it

01:05:45isn't always

01:05:46about you

01:05:47know getting

01:05:48to that run

01:05:49phase but

01:05:50let's start

01:05:51here and I

01:05:52think that's a

01:05:53very important

01:05:54point that you

01:05:54just made

01:05:55and yeah I

01:05:57think this

01:05:57call has really

01:05:58been fantastic

01:05:59and just opening

01:06:00our mind and

01:06:01also offering a

01:06:02little comfort to

01:06:02everyone out

01:06:03there because

01:06:03you just have

01:06:04to take the

01:06:05first step and

01:06:06that's that's

01:06:07all you have to

01:06:07worry about

01:06:08today.

01:06:14Any closing

01:06:15comments?

01:06:15Randall from

01:06:16your side?

01:06:18No I think

01:06:19I'm the only

01:06:20comment I would

01:06:20make is I'm

01:06:21excited about the

01:06:22future I think

01:06:22that we are

01:06:23certainly on the

01:06:25verge of a

01:06:26new decade of

01:06:29mortgage and

01:06:30those of us that

01:06:31have been in

01:06:31the industry

01:06:32long enough to

01:06:33watch the

01:06:34decades as

01:06:35they have

01:06:36evolved from

01:06:37pre-AUS days

01:06:39into AUS and

01:06:42now into as

01:06:44Joy mentioned

01:06:45imaged files and

01:06:47doing away with

01:06:48the paper and

01:06:48now we're in a

01:06:49space of how can

01:06:51we create

01:06:51efficiencies with

01:06:52AI.

01:06:53I'm very excited

01:06:55to watch the

01:06:56current and new

01:06:58generation of

01:06:58mortgage

01:06:59professionals

01:07:00embrace the

01:07:01technology and

01:07:02move forward

01:07:02hopefully cutting

01:07:04costs, reducing

01:07:05costs for

01:07:05borrowers and

01:07:07the industry

01:07:08alike.

01:07:11Absolutely.

01:07:12Thank you for

01:07:14that comment and

01:07:15Andrew thank you

01:07:16for that comment

01:07:16as well that yes

01:07:17we are in the

01:07:18crawl phase and

01:07:20those who are not

01:07:20crawling should

01:07:21first crawl before

01:07:23walking.

01:07:24Look what you

01:07:26have I mean you

01:07:27might want to

01:07:29deploy AI but

01:07:30maybe you don't

01:07:30have the data

01:07:31that is required

01:07:32for or you have

01:07:33the data but you

01:07:34don't have the

01:07:34ecosystem right you

01:07:36don't have APIs

01:07:37ready to run the

01:07:39AI models so

01:07:41there are a lot of

01:07:42permutation and

01:07:43combinations and

01:07:44that's why you start

01:07:45small start slow

01:07:46get the ROI get the

01:07:48confidence and then

01:07:49you move forward so

01:07:50that's the that's the

01:07:52approach and you

01:07:54know taking that into

01:07:55consideration I want

01:07:56to just quickly

01:07:56touch base on

01:07:58Sutherland's you

01:08:00know underwriting

01:08:01center of excellence

01:08:02is something that

01:08:03we've created you

01:08:05know a year back

01:08:06we've created a

01:08:07center of excellence

01:08:08which if you see

01:08:09the the top part of

01:08:12creating a learning

01:08:13center giving

01:08:15underwriters a

01:08:16refresher training

01:08:17every quarterly

01:08:19making sure that

01:08:20those who are

01:08:21working on FHA

01:08:22loans get trained

01:08:23on non-QAM

01:08:25other channels

01:08:26making sure that

01:08:27you uplift the

01:08:29knowledge of an

01:08:30underwriter or one

01:08:32of the top

01:08:32processes for them

01:08:33to become you

01:08:34know an underwriter

01:08:36or support

01:08:36underwriting so the

01:08:38top part technically

01:08:39focuses on in the

01:08:41human right enhancing

01:08:43them their knowledge

01:08:45you know when it

01:08:46comes to the product

01:08:47and if you see the

01:08:48bottom side of it

01:08:49it talks about tech

01:08:51it talks about

01:08:52using Sutherland's

01:08:54proprietary tools

01:08:55to for document

01:08:57processing for you

01:08:59know in-house tools

01:09:00for deploying bots

01:09:02or using AI powered

01:09:03knowledge libraries

01:09:04where you can type a

01:09:05question get a

01:09:07response from any

01:09:09guidelines you know

01:09:10from the agency

01:09:11internal external

01:09:12process flows so what

01:09:15we try to do is we

01:09:16try to get tech and

01:09:17human together right

01:09:19and this is the twin

01:09:20brother strategy that

01:09:21I was talking about

01:09:22that you need to

01:09:24have human right

01:09:25nobody is going to

01:09:27replace them they

01:09:28need to be there for

01:09:29a reason right a

01:09:31bot or an AI

01:09:32cannot take away two

01:09:34decades of underwriting

01:09:35experience from a

01:09:36human it's just going

01:09:37to follow a pattern

01:09:38a pattern that you

01:09:40are teaching the

01:09:41bot or an AI to

01:09:42follow you know

01:09:43like like the

01:09:44question the example

01:09:45that I gave that if

01:09:47there's a pattern

01:09:48that AI detects a

01:09:49particular segment or

01:09:51section of society

01:09:52and you know loans

01:09:54are in decline for

01:09:55them and if it sees

01:09:57similar pattern and

01:09:58start declining new

01:09:59borrowers right maybe

01:10:00we have to look and

01:10:02know what we are

01:10:02trying to teach the

01:10:03the bot or an AI to

01:10:05to come out with the

01:10:07output so it's very

01:10:08important that it's

01:10:09balanced and that's

01:10:10what we try to do

01:10:11with our center of

01:10:12excellence bringing the

01:10:14human technology

01:10:15together make sure

01:10:16they talk to each

01:10:17other they understand

01:10:18what technology is

01:10:19how it works why it

01:10:21should work is

01:10:23technically comes from

01:10:23them like what to

01:10:25Andrew's point is you

01:10:26involve processors

01:10:27underwriters closes

01:10:28try to understand what

01:10:29their challenges are

01:10:30that's what we try to

01:10:32go in the center of

01:10:33excellence which we

01:10:34have which is

01:10:35clearly given us the

01:10:36lift you know from

01:10:38an underwriting

01:10:38productivity

01:10:39perspective so with

01:10:40that I mean I would

01:10:41say it was a great

01:10:43learning for me right

01:10:45from I mean the

01:10:47panelists that we

01:10:48have I think is

01:10:50fantastic right we

01:10:52got to know

01:10:52Randall's view from a

01:10:54bank's perspective

01:10:55Andrew you know he's

01:10:57an AI expert right

01:10:59and then joy from a

01:11:01platform you know

01:11:03organization so all

01:11:04these perspectives I

01:11:05think have helped me

01:11:07and I'm sure those who

01:11:08are listening to help

01:11:09them as well

01:11:09thank you all so much

01:11:15for sharing all of

01:11:16those insights I do

01:11:17want to touch on a

01:11:18couple questions and I

01:11:19know that we discussed

01:11:21some of these answers

01:11:22briefly but I want to

01:11:23reiterate them at now

01:11:25that we're almost out of

01:11:26time for today so the

01:11:27first question that I

01:11:28have is looking back at

01:11:31how AI improves risk

01:11:34assessment compared to

01:11:35traditional underwriting

01:11:36methods

01:11:37I can touch base on

01:11:42that I mean if we look

01:11:43the traditional you

01:11:45know and Randall

01:11:46maybe you can help me

01:11:47here

01:11:47checking the income

01:11:50you know documents

01:11:52bank statements

01:11:53checking the DTI and

01:11:54then taking maybe

01:11:55maybe a decision but

01:11:56with AI there are

01:11:57different parameters

01:11:58that AI can throw

01:12:00it can look at the

01:12:01rent payment you know

01:12:02any utility payments

01:12:04that have been made

01:12:04get more insights into

01:12:06a borrower right

01:12:07which can help

01:12:08understand not just

01:12:10from an income or a

01:12:11DTI perspective but

01:12:12there are other

01:12:13perspectives from which

01:12:14an AI can help you

01:12:15know underwriters take

01:12:17that decision so that's

01:12:18clearly you know a

01:12:20differentiation that I

01:12:21see from what

01:12:22traditional underwriters

01:12:23would look at to what

01:12:25AI can do

01:12:26and what would a fully

01:12:29AI powered end-to-end

01:12:31mortgage approval

01:12:32process look like and

01:12:33is it desirable

01:12:35Joy I'd love your

01:12:37perspective on this

01:12:38so I do think that AI

01:12:41plays a part and I

01:12:42think that we can get

01:12:43caught up in that term

01:12:44AI but what we have

01:12:45found is when you dig

01:12:46deep and you're working

01:12:47with the client coming

01:12:49from a perspective of

01:12:50having a flexible

01:12:51mortgage tech stack is

01:12:54that we we ask first

01:12:55where do you want to

01:12:57insert that

01:12:58automation those

01:12:59validations hence my

01:13:01comment earlier about

01:13:02the excel file and

01:13:03learning well wait we're

01:13:04not ready to get rid of

01:13:05that we don't want to do

01:13:06that so it's like where

01:13:07do you want to insert

01:13:09that efficiency and

01:13:11typically it's it's

01:13:13different for every

01:13:14client but I think that

01:13:15there's absolutely room

01:13:17for us to improve I

01:13:18mean we have the

01:13:19automation we have that

01:13:20available there's a

01:13:22customer who wants to

01:13:23close the file in 10

01:13:24days and I'm like that

01:13:25there is no limitation

01:13:26on that outside of the

01:13:28regulations you can move

01:13:29as fast as as you can

01:13:31or as you want to but

01:13:33having the ability to

01:13:36put in the work to think

01:13:38through how you want

01:13:39that workflow to go and

01:13:41then take the time to

01:13:42help build it out it's

01:13:43really as long as you

01:13:44have a data-driven cloud

01:13:46based software sky's the

01:13:47limit you know you can

01:13:49accomplish what you're

01:13:49looking to accomplish

01:13:50I love that sky's the

01:13:53limit what a great

01:13:54place to wrap today's

01:13:56webinar thank you all

01:13:58so much for joining us

01:14:00today for smarter faster

01:14:02safer reimagining mortgage

01:14:04underwriting with AI a

01:14:06big thank you to our

01:14:07panelists Joy Andrew

01:14:10Randall and to Hitesh for

01:14:11leading us through today's

01:14:12conversation housing wire

01:14:15will be sending a

01:14:16recording out to all

01:14:17registrants you'll also be

01:14:18able to access the on

01:14:19demand version on our

01:14:20website at housingwire.com

01:14:22slash webinars thank you

01:14:24again for joining us we

01:14:25hope to see you all again

01:14:26soon thank you thank you

01:14:28everyone thank you

01:14:29thanks guys

Category

Transcript

Comments

Recommended