DN #25: The $239B Mortgage Wedge, 100% Accuracy AI & The On-Prem Comeback (w/ Danny Tang)

Danny Tang Founder of Tradata:Yes, ⁓ my pressure to hear. Yeah. ⁓ yes, ⁓ that's a great question because ⁓ I'm from Hong Kong and I'm just studying in the University of Hong Kong and ⁓ my major is ⁓ related to ⁓ some statistics math and finance and then I just ⁓ came up with an intern in the reinsurance group of America and the at the financial and I have spent more than four years in the financial industry in Hong Kong and I really love finance. Yeah. And ⁓ ⁓ the problem is that I just find out that there is a lot of document problem and then ⁓ they need to gather all the documents before making some decisions or some ⁓ managing the risk yeah afterwards but and ⁓ most of the most of the people they are doing the reading ⁓ the documents and I think it's ⁓ really time consuming and they should not be spending time here and this is spending on more than on judgment and after that I just ⁓ want to start a company in in San Francisco because it's really ⁓ a great environment for startup and then I just find ⁓ my ⁓ PvS colleague and they have some connections with the ⁓ US financial market there and ⁓ ⁓ I and I just interview all of them and I find out from all the interviews I find out ⁓ the biggest problems is that they has they have a lot of document processing ⁓ problems in the long qualified mortgage market. And so I just wanna solve that and just help ⁓ the non qualified mortgage industry to grow faster than ever because ⁓ is the right timing to scale it and I would say ⁓ normally I would say ⁓ the gate workers ⁓ especially for the creators and then ⁓ some entrepreneurs ⁓ the numbers are keeping re ⁓ I keep increasing ⁓ for ⁓ the next few years or next ten years I would say and they may need to buy home. and the real estate investor want to invest in homes and then just ⁓ want to ⁓ lease it and then just keep ⁓ keeping the cash flow and I would say so we just want to help ⁓ to deal with this problem and ⁓ without adding the head count because I really know ⁓ the the document reading task I would say is really frustrating for ⁓ long cumulative underwriters and We really want to help solve it. Yeah. ⁓ yes. ⁓ the for the market size I would say ⁓ for ⁓ two hundred ⁓ for for two thousand and twenty five ⁓ the market size is about two hundred and thirty nine billions for a long qualifying mortgage and which is nine percent of the total ⁓ US mortgage market. And the market size is really large and it's really keep increasing, but I know that ⁓ this problem ⁓ is really ⁓ I would say not obvious for ⁓ many founders and I would say I love to solve ⁓ those problems that no one cares but is really serious and really keep increasing and then I I want to be ⁓ the first mortgage tech company that really help accelerate the lenders and ⁓ building trust between them because I we know that ⁓ the mortgage landers they have s ⁓ the cons ⁓ I would say ⁓ for the mindset they are just conservative to AI because AI cannot ⁓ doing everything one hundred percent correct and ⁓ they fear that they ⁓ replacing by AI but I would say our AI is not ⁓ replacing any underwriters. We're just helping them to do everything faster and ⁓ to our AI is not ⁓ replacing any underwriters. We're just helping them ⁓ to do everything faster do the boring job, yeah, for just reading the document and then remain the most important task for ⁓ making decisions to them. Yeah. Yes. Yes, we are ⁓ getting to one hundred percent because it's ⁓ it's really building trust between ⁓ our AI and ⁓ lenders and ⁓ LM is not accurate enough for those ⁓ for the regulated financial industry because they really need the one hundred percent and if there is any errors the investor would not buy their loans and I would say ⁓ they need to carry on ⁓ the ⁓ the buyback risk and is ⁓ the loss is unlimited and they may be sued by SEC and yeah and we are just building this specifically because for LM they are just ⁓ the problemistic ⁓ machine I would say and ⁓ And they may have some errors and they have hallucinations and ⁓ they are not specific enough for this regulated market because they ⁓ don't have much data about it. Because ⁓ all the banking industry or the mortgage industry they will never leak out their data to AI. And so they are just keeping their data in their database. And their ⁓ local drive, and we're just doing all the specific tasks for them. Yes, yes. Well we are using APIs, yeah. So just running it ⁓ in the local drive and then ⁓ yeah privately ⁓ because ⁓ we really care about the trucks and ⁓ the privacy ⁓ for the regulated financial industry and yeah. Yeah we care a lot about on this. Yeah. ⁓ we just interviewed ⁓ more than ⁓ sixty ⁓ underwriters and then just ⁓ know about ⁓ their workflow and ⁓ really ⁓ understand about the problem is and just help to solve it by ⁓ creating our demo and ⁓ for the last few days ⁓ we have ⁓ a potential customers and ⁓ but I I was not say it's customers because we are just launching our closed beta program ⁓ to our partners and just doing the design partner phase and yeah and then we can just keep refining our demo so that we can just ⁓ launching our pilot later and ⁓ maybe ⁓ I would say our target is the last quarter of this year and then just ⁓ get the first customer and yeah. ⁓ yeah. because for the long workflow is that it's it's like ⁓ the underwriters are ⁓ doing ⁓ the ⁓ reading the documents like ⁓ the four hundred page of band statement and then ⁓ another tape would be ⁓ doing the Excel calculation for calculating the qualifying income and then ⁓ they would just put ⁓ the data back to L O S and then just approve the loan. or not and yeah and the process may be spending almost ⁓ more than one hour per file and most of the time are spending on the ⁓ document processing ⁓ work and ⁓ I just hear ⁓ one underwriters with more than twenty years experience that and he tells me that ⁓ even if he wants to approve the loan And ⁓ he needs to read all the documents, get the all the test return, gather all the ⁓ PNL and then gather all the WTOS and then together all the documents and then to prove ⁓ true stories to ⁓ approve the loan so that they can tell this story right to the investor. Unless the investor will not buy back the loan and and yes, and that's it. Yeah. ⁓ yes, that's a great question. Because ⁓ just like I mentioned and I really want to solve some problems that ⁓ no one really actually know about with but it's really becoming more serious and more ⁓ increasing for that time and I can see the trend because ⁓ because ⁓ the numbers of ⁓ the creators like the youtubers and ⁓ the Instagram and the ex and then everything for creators and all the gate workers and then doing some work remotely and ⁓ some entrepreneurs ⁓ just like us. Yeah and then ⁓ the numbers would be keep increasing because of AI because they can ⁓ easily to do all the creators job and then creating all the AI ⁓ image, AI ⁓ views and then ⁓ they'll just keep increasing and I we and we can see that and they need a home. Yeah, and then they need a home. And also for the real estate investors, they will keep ⁓ investing the homes and then just lease them out to earn the cash flow and ⁓ the trend is keep increasing. ⁓ but ⁓ the headcount I would say is not ready for this increasing trend. And we are just ⁓ helping the underwriters to scale the loan without sacrificing all the buyback risk because ⁓ the current AI underwriting they're really scared of using it because ⁓ they cannot ⁓ give one hundred percent accuracy and they may have some data leakage and they ⁓ they are not really solving their problems. They are trying to replacing the underwriters. And they are really scared of it. And what we are different is that we are just helping them to ⁓ scale the loan without adding the head counts and accelerate the loans ⁓ with ⁓ save and ⁓ adhering all the investor guidelines and then just help them do easier do things easier. Yeah. Yes. ⁓ yes, ⁓ because ⁓ just like I mentioned before because I work in in the financial industry before and I really see that problem for the document processing so I just try to build some side projects and ⁓ and and then ⁓ building it to to be open source ⁓ so that I can learn from it and then and ⁓ I just try because ⁓ I'm just studying the degrees that is not included in the computer science just a little bit just basic for py Python I would say just basic and ⁓ I want to self learn some codings because ⁓ some AIs ⁓ are are are not ⁓ allowed to use here. So I I need to ⁓ learn itself by myself and then I just ⁓ just try doing and just try building something by learning and then yeah. And then what I learned is that ⁓ the first thing is that ⁓ if I just learning from all the books and then I I I cannot just learn learn that stuff. Yeah. I re I really need to learn by doing. Yeah. Just trying to build in something real and for a document process like ⁓ the resume and ⁓ the receipt yeah, receipt reading and ⁓ I just I I c I will just remember I just achieved ninety five percent accuracy for that time and yeah, I and I'm trying to improve them to about one hundred percent right now and I would say yes just try and learn by doing. Yeah, and then just keep improving and refining all the stuff and then I just try using all the skills, ⁓ put it back to Tradata. Yeah. Yes, yes, we are building the native one. Yeah. I think ⁓ a little bit different. Yeah. The marketing will be a little bit different because ⁓ I know that ⁓ the B do B SAS is normally just ⁓ building a plot plate and then and then just used by anyone but ⁓ but What I've learned in working in the financial industry is that they really want a PiFec AI software for them because ⁓ they really sensitive of any data leakage. Yeah, just like ⁓ the latest news is that ⁓ when the clock just released is new is the latest version and then all the bands just ⁓ just to be careful and then to prevent any ⁓ security risk. Yeah, because financial is really ⁓ the foundation of of a country. Yeah, and we serve as the capital, I say just like the heart of the capital. And it's really important. So the data must be reserved within themselves instead of leaking out. But ⁓ most of the software like ⁓ the Microsoft and then ⁓ it's okay to use the things like that because all the data is not ⁓ really important when using for like ⁓ Excel to do the calculation. But the really data and ⁓ the some documents and some ⁓ Pi V C datas is really solved in some ⁓ their database and it never want to ⁓ give us to anyone to know, yeah. And so we're just building the Python AI and then ⁓ just making something what they want. Yeah. ⁓ first thing I would say the trend would keep increasing and then ⁓ for now ⁓ non qualifying mortgage is just nine percent of the total mortgage and I think I think ⁓ in the next ten years it would be more than thirty percent. I would say. Thirty percent because ⁓ the number of entrepreneurs is keep increasing just like ⁓ we are doing the start up and then everything everyone is now doing startup right now is AI startup and ⁓ they need to have a house. Yeah, they may ⁓ rent a house or buy a house and then it also involve in ⁓ the non qualified mortgage and then just directly or indirectly the first picture which would be like that and the second picture would be I would say ⁓ they would just ⁓ tends to having a long cloud software instead of cloud software, cloud native software, because for other industry they may ⁓ they may be helpful for using the cloud but ⁓ for financial industry I think ⁓ it's really risky for using cloud native software because just like ⁓ what cloud said yeah the cloud said the fable ⁓ the the media Mevo yeah the Mevo is really strong for the model and then they may ⁓ raise some security risk and they may ⁓ some hackers may use their models to hack into the financial system of ⁓ different countries and then ⁓ and I think cloud native software would be very dangerous for ⁓ this risk and ⁓ we are just doing the opposite things to keep our financial industry and then ⁓ that's a second picture and And the first picture would be ⁓ some ⁓ landers giant like A and D Mortgage, they are now building the ⁓ home ground no ⁓ home ground tools for ⁓ building their AI for themselves because they are having a lot of budget to hire a software engineering team and then just build the AIs ⁓ themselves and ⁓ for the some ⁓ small to medium lenders they don't have the budget to do that. And ⁓ it would be a disadvantage for those small to medium without using AI because the more ⁓ the mortgage ⁓ brokers will just give the loans to the faster lenders in the instead instead of ⁓ some ⁓ lenders without using AI. And the gap between the giants and the small to medium lenders would be widened but ⁓ we are here to solve it. We are here to narrow down the gap because ⁓ we can just provide some ⁓ our software to make the small to medium lenders and then to use AI and then use our AI and then ⁓ just scale the loan prevent ⁓ prevent the loss to the s giants lenders and yeah. And I would say the competitive mo ⁓ I would say ⁓ the long qualified mortgage market would be more competitive than before. Yeah. Yeah. And the gap would be narrow down. I will say ⁓ if without us, I would say ⁓ most of the gi most of the land giants would just acquire some smaller lenders and then just make the market ⁓ be more monopoly. But ⁓ after having us, okay, we were just ⁓ helping the small to medium lenders to use their ⁓ you to use our AI and then to scout the loan following faster and ⁓ and the loan volume would be ⁓ just very close to the lander giant. And then the market would be ⁓ more competitive b than before. Yeah, I would say. Yeah. Yes. Yeah. Yeah. Yeah, yeah. The gap would be ⁓ getting closer because ⁓ we can just help the small to medium lenders and then just ⁓ scale the loans and then ⁓ I would say ⁓ the most important thing I would say is that ⁓ There will be more more companies that ⁓ just building something ⁓ similar to us, I think in the future because they may they may see us and then ⁓ they can just ⁓ solve the problems with those ⁓ small to medium landers or landers giant because ⁓ I would say ⁓ their software def ⁓ they have budget for hiring ⁓ a strong team of software engineer but the software engineer may not ⁓ know about their problem and ⁓ they may not have ⁓ some experience in the financial industry and then just they may just build according to what ⁓ the CEO said and then ⁓ and then just follow all the instructions but they may not know ⁓ the problem well. But we are loading the problem well so ⁓ I would say ⁓ the gap would be ⁓ closer, yeah, in the future. ⁓ yeah. I would say the most ⁓ fun things would be I really enjoy ⁓ doing anything related to finance because I always read in all the books related to ⁓ like ⁓ Rich Dai Puddha and ⁓ Ray Dalio and Warren Buffett and then I always read all the financial books and then learn about the financial financial industry since I was a kid and I really enjoy to solve ⁓ the big problems in the financial industry and I think for this problem ⁓ just like I mentioned before and ⁓ it's no one cares about it and no one's know ⁓ deep about this but it's really getting serious and ⁓ it has a really potential to be ⁓ ⁓ a a a real problem real urgent problem to solve and the one who solved it well and really fast would be would become a real giant ⁓ technology company I would say. And I really enjoyed chatting with all the financial industry ⁓ colleagues and all the underwriters and yeah I really enjoy my job and the thunder yeah Because I love finance. Yeah. Yeah, thank you, Nexus, for having me here and yeah.