00:00I'm Alison LaBorgia with HousingWire and today I'm so excited to talk to Dr.
00:08Olav Laudy from Tevant. How are you today?
00:10Very well, thank you. Thank you very much.
00:13So I am excited to talk to you and let's start with your journey into mortgage technology and
00:21how you became focused on AI driven solutions for mortgage.
00:25Right. So I think I am a traditionally machine learning guy. So then naturally you get into AI,
00:36right? When machine learning transferred into the AI domain. I joined Tevant and Tevant has a mortgage
00:43arm. And naturally I was curious on how the things were already automated and how I could help
00:51automate that more using AI. Now, when it comes to AI and mortgage specifically,
00:57our industry has a lot of hype. So I'm really interested in hearing how it's different for
01:04Tevant and their approach to AI implementation and how Tevant has strategic solutions to solve
01:14real friction points when looking at AI technology. Right. So I think the key is
01:21understand the mortgage process as professionals act, right? And we deploy a lot of actual mortgage
01:30professionals, people with business backgrounds in mortgage, and they understand what the friction
01:35points are. So when I propose something, they can say yay or nay to it. And that allows me to develop in
01:41the right direction. But other than that, from an AI point of view, we're never deploying AI for the
01:49sake of AI. You need to deploy AI because it solves indeed what you correctly mentioned, a friction point.
01:56Anything that takes a lot of time for someone can be potentially enhanced by an AI feature. We're not
02:03looking to replace underwriters or lenders in entirety. We're looking to enhance the process so that those
02:14people can do their work better, faster, smarter. That is very interesting. So it sounds like there's
02:20really a meeting of the minds in Tevant's approach to AI. You have a full understanding of the process,
02:27and then you have a subject matter expert like you who says, okay, where can we alleviate some
02:33of these pain points to make the people better? So I'm really excited to dig into that a little bit
02:38more. Can you share an example where AI streamlines tasks for the loan officers or for the underwriters
02:46in this process? Yeah. So here's a very simple example, right? An underwriter typically is asked by a
02:54loan officer, hey, can you tell me the status of the loan, right? So that's a very common question.
03:00So what does the loan officer, what does the underwriter need to do? They go into the loan,
03:04they kind of look what is missing, what is the status, where we are, how much still need to be done,
03:13and then that is one task. And then they have to kind of all write that up and send it to the loan
03:20officer. So a simple email like that will take, if done properly, 20 minutes. If you take three
03:26minutes for it, you're basically just saying like, okay, it's fine, don't worry about it, right? But
03:32that is also not fair to the loan officer. So an AI, an agent that is aware of the loan can do that
03:40searching, right? So we have built this, and it is an AI set that's called chat with loan, right? And you
03:50you go in there and say like, write a note for the loan officer. The AI reviews all the elements in
03:58the loan, what is missing, what's the status, what is the time to close, etc. And then writes that email
04:05as an update for the loan officer, the underwriter, and this is a very important point, looks through
04:12that email, says like, this is right, or this is not right, this is what I do want to say or not,
04:17it has the ability to update something. So this is the human in the loop. And once done says click send.
04:24And that process now in its entirety will take three minutes instead of the 20 minutes. But in those
04:31three minutes, a complete email is actually being compiled as something that is relevant
04:38and useful for the loan officer. And that's just across one file, which is fantastic. So if you can
04:43save just, I mean, what sounds like upwards of you're cutting 75% of the time out of it for multiple
04:53files, it sounds like a massive time saving, which is fantastic for everybody who's involved in
04:59processing that loan. So I'm really curious if you could tell me about how this extends over to the
05:06borrower experience, especially for consumers who we know are now remote, time constraint, or even
05:13managing a complex financial situation. Right. So we have our mortgage avatar called Maya. And Maya
05:22basically accompanies you from the start of your loan all the way till the end and into the servicing
05:31end of it. So when you start your mortgage application, Maya greets you and will basically introduce
05:39herself, letting you know what she and what she can do. Right. What she can do is very clear. She's not allowed to
05:47say like, Oh, I can give you this mortgage for X percent of interest rate. But what she can do
05:53is help you fill out forms, answer questions along the way, and do the important checks. And so for
06:01example, if you need to upload a 1040 tax document, and you upload a different tax document or a different
06:09year or something else, right, it is immediate immediately noticed. And then that gets fed back
06:17to you. So like, well, this is not the right document. Now, if you compare this to the traditional way,
06:23normally, you would upload the document. And then that goes to a loan officer, either the loan officer
06:29sees that or not goes to the underwriter, then that goes, the underwriter says, yeah, but this is not
06:34what I need. And now this whole hop needs to hop back. And that gets two, three days down back to
06:40the borrower, the borrower lives in anxiety thinking like, Oh, well, I get my loan or not. Or even worse,
06:46has a parallel track with another lender is exploring that. And maybe if that goes faster, he will drop you.
06:52Right. So the here you see an avatar that is truly functional in doing things for you at the moment that
07:01you need it. Another example is the clearing of conditions after they get access to your your bank
07:08transactions. For example, this is a very common hurdle for people, they need to write a gift letter,
07:15or they need to explain some transactions. Well, if an avatar kind of scans your bank transactions,
07:21and says like, hey, already explain me this because I because I know that we're going to get answers
07:27or questions around this, you just save valuable time for both the borrower as well as the underwriter.
07:34That sounds like some really fantastic improvements that are now enabled by utilizing AI in the borrower
07:41experience. However, there the industry at large has some concerns around compliance, safety and
07:48usability. How would you address those concerns? Right. So I think the compliance is
07:54if you compare, if you would replace the word AI by computer, right, and you would kind of realize how
08:08people are freaking out over all compliance, what what happens if you have an AI do this, right? What
08:14happens if you have a computer do this, as compared to the previous process where people would write
08:20everything out. So obviously, it is not the computer who decides who gets alone or not, but the computer
08:28makes the whole process very smooth, and frictionless. And so with AI in place, that is an even further
08:38enhancement of that process. So you always have a human in the loop. And you always have things tested,
08:46and make sure that nothing goes wrong here. So I think the human in the loop, as a final approver,
08:56is key to this whole process, and not thinking that you just hand it off to a big machine, and then
09:03you hope for the best. The other point is obviously, like, you do deep user research, right, experiments to
09:13make sure we're actually conducting one at the moment, to see if people would like to see an avatar on
09:20screen, or they look like would like to see a speech avatar. So you kind of see some some waves on the
09:28screen, if the avatar speaking, and then how that functions, if they do have questions for this avatar,
09:35if they feel that the avatar can can help them in that process or not. Or would they, for example,
09:42rather type something in, right, it can well be that for for certain groups of users or borrowers,
09:51they would rather kind of ask a question and not have a like a presence on the screen. Personally,
09:58when when I looked at this, I find an avatar at the screen, and filling out forms with me and for me
10:06is lovely, right? So I if I can just talk to my computer, I say, Okay, can you fill this out? Can
10:11you fill this out? Or if I am not entirely sure what I need to fill out, I can ask questions, or I can
10:18describe what I want. And the avatar smart enough to take from what I described, and then fill the form for
10:24me, I think that is the next level of interaction. But we're testing that out as we speak, to see what
10:30is the best way and even let people choose their their mode of interaction with this process.
10:38You just mentioned a couple features that you're looking at implementing and testing. How do you see
10:45AI continuing to transform mortgage operations and borrower engagement? And what innovations are you most
10:52excited about? Well, we're currently, apart from the voice interaction, which just matured, I guess,
11:00in the last half year, we're exploring something that we refer to as policy as code. And so traditionally,
11:08you would have, like all your guidelines. And if you were to automate that, you would have to translate
11:14them into a computer readable rule set. And that rule set can be used in a scoring or a rule engine to test
11:24against, like the actual mortgage details of a borrower. When you have something like policy as code,
11:36you're just kind of omitting that complete step, right? You have the borrower's details, and you have
11:43your guidelines, and you have an AI that very smartly goes through this. Now, obviously, you cannot,
11:52if you try to kind of think about this, it is not that you have like a chat GPT, and you provide your complete,
12:01say, guideline document or guidelines documents, right, into chat GPT. And you add all your borrower
12:09information, and you ask the AI, please figure this out. That's not how it works. You still need to kind of
12:16cut things up in manageable bite size chunks, and then ask that to the AI. But you'd be surprised how strong
12:24AIs are today, to take very complex rules, and then apply certain conditions to that, and data to that,
12:33to see if things work out or not. In fact, so much so, that if I try this out, and I'm working with those
12:43complex rule sets, that I will need to really sit down and five minutes kind of compare details to
12:49understand the complex conditions, whether they apply or not. And you provide that to an agent,
12:55and the agent says, okay, here you are, this is the outcome. And so I think that will lead to
13:03much faster processing, as well as much more concise compliance.
13:09Dr. Elaudi, thank you so much for walking us through Tevant's approach to strategic AI
13:16implementation in mortgage. It sounds like there's so much that you guys have accomplished in such a
13:22short period of time. I'm so excited to see what the future brings. Thank you for joining me today.
