Weekly AI Update: Why Most AI Strategies Fail—and What Real AI Looks Like
Artificial intelligence is no longer just about deploying the latest chatbot—it’s about building solutions that can be trusted in high-stakes environments. In this week’s AI Update, David Lykken sits down with Pavan Agarwal and Jennifer to explore the critical difference between organizations that merely claim to have an AI strategy and those that have built one around proprietary, purpose-built intelligence. From the emergence of “PropAI” to the role of human oversight, warranted AI decisions, and dramatically reducing operational costs, this conversation examines how specialized AI is transforming mortgage lending and why the future belongs to AI systems designed for accountability, accuracy, and real-world outcomes rather than hype.
[David] Folks, we’re here with another AI update with Pavan and Jennifer. Good to have both of you here. I want to get into two things today, one of the wanna talk about, well, actually three things. I want to start with AI strategy, what is the difference between someone who has an AI strategy and someone who thinks they have? I mean, there’s a lot of people going about saying I have an AI strategy, and there’s other people that really do. What’s the difference, Pavan? Start it off.
[Pavan] Well, I’m going to start off by first introducing a brand new term to your audience. And it’s called Prop AI, P-R-O-P-A-I and prop does not stand for property. It stands for proprietary, so I’m going to be in about two weeks at the AI4 conference. So you all check it out. AI4.io is the largest AI conference of the year. It’s in Vegas. Some 10,000 people will be attending. I’m one of the keynote speakers. And all the buzz this year is about Prop.ai because
[David] So explain prop it stands for again, proprietary.
[Pavan] Proprietary, right, and are customized or problem specific, AI to solve specific problems. That’s Prop.AI. Okay.
[David] Now does proprietary mean that can it fall in the lar LLM category where it’s a public or using public data? Or is it is private as well as proprietary? What does proprietary imply private?
[Pavan] So proprietary means that it is not using an LLM to make its decisions.
[David] Yeah, by the way, LLM stands a large language model for those that are listening or not really but not l don’t have a good knowledge yet of this whole AI experience. So okay, so say that again.
[Pavan] Yeah. So proprietary AI is an AI that does not use something like ChatGPT or Google Gemini or some other model like that, an LLM model, to be its brain. It’s a customized neural net, it’s customized AI to solve specific problems. So for example, the most commonly unknown and used prop AI out there is the AI in your test, your self-driving AI. That’s a prop AI. That’s not Grok. That’s not Elon Musk’s Grok that’s running a Tesla. No, that’s a customized AI that can make microsecond life and death decisions. And it does it all the time, all day long and it runs completely in your car by itself. And so fully autonomous. I mean, anybody who owns a Tesla, you all know how well it works. It’s not perfect, but it’s really close to being perfect.
[David] And they keep perfecting it. Just yeah, it it just really
[Pavan] And it keeps getting better, right?
[David] It’ll even drive you to a charging station. You say, I want to go to Kansas City if you’re you know in Dallas and Jay driving to Kansas City, it’ll even when in time it knows when to it needs to hook up, it’ll drive to a charging station. And I mean you get you do still have to get out and plug it in. I guess that part isn’t automated. But it’s astounding that it’ll go there and park itself and say, Okay, driver, you’re time to wake up and go to work. Go plug it in.
[Pavan] Yep. now Toyota’s new solid state batteries will go a thousand kilometers. So you don’t even have to charge it anymore. But I digress. I digress.
[David] Wow. I did not, you know, they’re the king of batteries. So, Jennifer Pavan is the guy that came up with the design for this prop AI. You have been integral in this, and I love how Pavan honors you. And he talks about you being the smartest person in the room all the time. It’s just a unique partnership between the two of you. And I want to get into strategy because a lot of people say that we have a strategy of an AI strategy for our company? You gave a good example a while ago, about a banker that was visiting you that was touting their AI strategy. Tell us that story.
[Jennifer] Sure, I mean Pavan is just being humble to call me the smartest person in the room, but we all know who the smartest person is, so
[Pavan] Now, wait a minute. Wait a minute. Wait a minute. What makes me smart is I have smarter people working for me.
[Jennifer] All right, we could keep going on on that discussion forever, but let’s get to
[David] Okay.
[Jennifer] David’s question here, which is you’re right. like we had a banker visiting us yesterday and we were talking to him about angel AI and the word that he heard from us was the warranty. We’ve already always talked about angel AI’s responses are warranted response. He says, Hold on, hold on, what are you telling me? Are you telling me if you get a response from Angel AI, you’re putting your money behind it? I say, Yeah. And so at that time he told me this example where his company in this AI initiative have established co-pilot for all their employees.
[David] Microsoft Copilot, yeah.
[Jennifer] And now he goes to the co-pilot, and there’s a data that that co-pilot sent him with some information. We don’t have to go into the details. So all he did was ask the co-pilot, hey, you sent me this data. Tell me something about this information that you have in this file in this particular area. He was very specific about it. But the co-pilot kept telling him, I have no idea what you’re talking about. I have no idea what you’re talking about. And then he went one step further and said, You know what? Okay, tell me this. is this a good time for me to go and cut my trees in my lawn? And said, it’s a perfect day for you to go out and cut your trees. So he goes outside and guess what he learns? A neighbor 10 houses down the street got hit by a lightning. And he runs back inside and he uses words that I can’t use here to the to the copilot. It’s like, what the are you doing here? You told me I’m okay to go and you know, and and
[Pavan] That’s not an exaggeration, that, Jen?
[Jennifer] It is not an exaggeration. I wish I could have recorded and played here. In fact I’m not even saying the words that he used when he was talking with us and then he goes back and asks the same copilot. It’s like hey you said this but guess what? This is what yeah, it is raining. Hell, I know it is raining outside because I just went outside, but you just told me it was a good day to cut trees, right? That is what large language models do. And that is the difference between a prop AI, which is an industrial strength AI, compared to an LLM, because all these LLMs have a disclaimer. Don’t trust my answer. Whereas Angel AI or others give you that warranty.
[Pavan] Yeah, and there was actually a news story that came out yesterday where Chat GPT talked a lady into killing herself. We walked in front of a bus. So this is a real problem because it’s model sycophancy. Start reinforcing what you want to And they start creating this feedback loop. So if you’re expecting a sort of, if you go down a dark path with the model, it’ll just keep reinforcing that.
[David] Well, I I think that’s what a lot of people I get concerned about executives. I’m thinking one particular executive here in the industry is talking about how he spent $200 million developing a developing AI and his business is running on AI. that it, you know, they say hope is in a strategy or prayers in a strategy. I mean, it there’s AI has got to take some design. You’ve been working on this and designing this as a result of your dad’s inspiration or is isn’t it was it inspiration or him just driving on you
[Pavan] It is basically, you know, Asian parents saying, go do this. Yeah, orders, order number one, get straight A’s. Order number two, go automate this industry.
[David] I knew your dad so well. I have much respect for him. And but he was, I mean, he was delightful to work with, but he was a mentor. We could go on and on about that. Someday we’ll do a whole podcast just about your dad and the and the things he taught us in our industry, those that were got to know him really well. And it was such a blessing. But he did drive you to create a design, Pavan, but did you have like a great design or what was behind? Why do why did you create this? Why did you create Angel AI? What’s your why?
[Pavan] Yeah, so, you know, going back to when I was a kid, I mean, his instructions were, hey, just look at this process. It’s just data, it’s just decisions. Computers make decisions. So why can’t a computer do the whole process? and that was the direction and that’s what we started building. That’s what I started building. And, you know, it took that long, 40 years to trial and error, trying different things. You know, I wrote my first AI module. I wrote it back in like 1985. And then in engineering school, I took a bunch of machine learning classes and neural net designs and so forth. I’m going off a deep end here. But anyways, in the bottom line is that if you have robotic manufacturing in the service sector, and the mortgage industry is a service sector and what we’re manufacturing here is a delivery of a service and this is a complex service that has lots of decisions that have to be made and a lot of data that has to be processed to deliver the service accurately. And it’s a high stakes service in the sense that there’s a lot of money involved. And you make a mistake, you could wire the wrong amount of money, you could wire the money to the wrong party or you could approve a loan that shouldn’t be approved, or even worse, and this is what happens in the industry all day long, is loans that should have been approved get declined. Because the actual numbers, the seven years of reviewing national HMDA data, is that one out of two black borrowers that are declined should have been approved. Two out of three veterans that are declined should have been approved. Three out of eight Hispanics that are defined should have been approved. So these are terrible, terrible numbers and if you have an AI doing it, which is what we’ve been doing for the past since 2019, we have the data, our data shows that it’s almost equal, the possible, right? Because that human guesswork disparity, you know, how the underwriter feels that morning, how the loan officer feels that morning, how well the how well it’s packaged, how the exceptions are dealt with, all of that stuff. And most importantly, what we see in this industry where a lot of the expenses are lost is the layers of sales and operations managers at least 50 basis points is lost and manager and another manager, and then you have people who check and then other people who check the checkers, you have layers of QC.
[David] You’re getting into Yeah, you’re getting into cost saving. I mean, there is a is clearly a cost saving, but Jennifer, he talked about the people that are being excluded from being getting mortgage loans. But I’m thinking of the pain of how many stories have you and I listened to, Jennifer, over the years where someone’s sitting at the closing table and they get the phone call. Oops, we messed up something. We were looking at your income and now they can’t close for whatever reason. That is I mean, that is reputational risk beyond belief, legal risk beyond belief. There’s just so many things. And this solves that. Something you explained, you don’t guarantee, but you warrant. Explain the difference between the two, Jennifer. Let’s start with you and then Pavan can add to it.
[Jennifer] I mean, I want to say guarantee, it’s just a legal compliance thing that I can’t use the word guarantee, but it’s really a guarantee, but I have to just use the word warranty, right? The reason I jokes aside, the reason we don’t warranty as against a guarantee, because we all know that sometimes as a consumer myself could get excited while I’m applying for a loan. I’ll go shopping and I’ll go get a car. Right? Last minute changes could happen, right? So now the loan that I may qualify for, I couldn’t qualify for because I added on another debt, right? So we want to make sure that the consumers understand their part of it as well in this transaction, right? It’s a two-way trust, right? They have to trust us as a lender to keep up our commitment and lender has to be able to trust the borrowers that you are you have told me that these are the debts that you have, and there’s nothing more, nothing new that’s going to come up between now and I fund the loan, right? So that two-way trust has to be established and since the other side is the partner in this relationship, it’s called warranty. Right? I can’t fund the loan no matter what you as a consumer do. There are certain responsibilities that has to be met from your end as well. That’s the difference between a warranty and a guarantee.
[Pavan] Right. Like for example, to underwrite a file, you need 30 days pay stub. I approved it conditioned to 30 day pay stub, but if you don’t give me the pay stub, I can’t fund the loans. So that’s why it’s a warranty.
[David] One that gets into this is a lot has to do with communication. It’s communication with the buyers, making sure things are clear. That’s one of the things that I like about the system. But part of your AI strategy, Pavan, which I think is brilliant, is you have defined where computers should be communicating and where humans should be communicating. You used to call Angelistas. I think you’re giving them now a new name. I’m hearing you’re recalling them something new, but I love the name. But that’s where the human factor. So it’s a partnership between human and the co and literally technology. Yeah.
[Pavan] Yeah, yeah. So I’ll talk about that. No, we still call them Angelistas. There’s three layers. There’s Angelistas, then there’s Angel Advisors, and there’s Account Executives. So the human component is so that the AI will email with chat with the customers, whether it’s the consumer, the title company, or the mortgage broker, or the realtor, right? It will do its best to interact. there’s, you know, in this business, so much at stake, so much stress everyone has in this business, you’ll need a human touch. Not all files, but some files you need a human touch. That’s where the Angelistas come in. They’re the first line. And then from there, it gets escalated to an angel advisor who are much more in-depth, knowledgeable about the whole process and then above that is the account executives who are complete masters of the domain.
[David] Very good. Okay. So do you you but the point is in your strategy, your AI tech strategy, you have humans in the loop. That’s very, very important. And I think that’s encouraging for many people who are thinking AI is going to put us out of business. It’s going to realign business tasks, who’s doing them. But there’s a clear human element that’s needed in here yet. And will that will that ever change?
[Pavan] The human element is needed in the relationship building. It’s still, you need a human to talk to the customer and encourage them, help them understand that there is actually, it’s not just a machine, there’s people that care behind it. And so you can’t take away the care piece of any business, especially in a service business. So all of our hiring is in that part of the business higher in the back end. We don’t hire for underwriters. And we have zero funders and zero loan processors because AI does everything. So the actual manufacturing of putting, bolting on pieces together and producing a final product and painting and polishing it and all that stuff is all done by robots. But the communication, to produce that end product, we may need some input from the customer, that communication, a human will step in and make sure the customer is feeling good about the process. as watching its deal get manufactured, you can see it along the way properly.
[Jennifer] Introducing the human itself is actually done by the technology itself because as you know, Angel AI is an empathetic technology, right? She understands when that human needs to get involved and she will actually include that human as soon as she understands that hey, I need a human here, and she’ll connect them with the consumer immediately. So it goes back to the it is intelligence, yes.
[David] And that’s not a weakness. That’s intelligence. And I think that’s the part that a lot of people, I mean, I’ve heard people criticize. Well, some of these companies, they have all their, they have they say they have a veneer of they say they have an AI company, but then there’s a veneer, but there’s hundreds, if not thousands, working in India behind the scenes to support this. That is just there’s humans evolved at the right place at the right time. But this is what’s the most poignant about having a clear AI strategy is the cost to originate. Now, folks, you’re going to hear a number that Jennifer’s going to give us here a minute. This is not hyperbole. This is not made up. This is real. Jennifer, as a result of Angel AI, the cost originate a loan operationally. I’m telling you, back out the compensation, I mean the loan originator compensation, backing that out. The operational cost for Sun West to originate a loan is how much? And I wish I could do a drum roll in the background.
[Jennifer] It’s okay. It’s a hundred and twenty five dollars or less.
[David] $125 where the industry average, all in, it’s $12,000, 11, $12,000, $13,000, depending on the company. Back out 60% of that for origination costs. So it’s still people are running, you know, four, five, six thousand dollars for operation. And you got it down to 125. And there’s still people involved. That is staggering. Pavan, I don’t know what you had in your brain when you started this thing, how to design it, but did you have any idea you could cut the cost to that level?
[Pavan] My wife says she doesn’t always say it my brain either, but that’s okay.
[David] Yeah. yeah, that’s truth. I remember John Macnell’s wife said one time, she says, I can’t believe what you say on stage up there. It’s just I can’t believe what comes out of your mouth. If you heard what’s going on in my brain, you can’t believe what doesn’t come out of my mouth. So I think there’s a lot of us have a lot of things going on in our brain. Some of it’s probably, Pavan, did you have any idea that you’re going to be able to get it down to that low of a cost?
[Pavan] I knew it was going to be low because it’s like, how much does it cost when you have Excel spreadsheet and you change the number and it recalculates 10,000 rows and gives you the number at the bottom, right? You don’t even think about the cost because it’s like, if you did it by hand, it would take you hours, right? But that’s why we have technology, that’s why we have Excel spreadsheets. Now imagine that, okay, apply to your whole mortgage process. And my thing to Jen is, is why is it still at 125? Why isn’t it at 25?
[David] Yeah, yeah. We we have Hari incarnate here, ’cause that’s what Hari used to do with them on all the time.
[Jennifer] It’s more than Harry and Heart, which is true as well. But that’s what makes us fun, right? developing a product and creating engineering is about having fun, right? And this is what
[David] Yeah. I’m glad you find that function. Yeah. Well, it’s and I know you say that w sincerely, and but it is always your there’s it’s it’s always that attitude of Kaizan, it’s never good enough. We can do better, we can do better, we can make it better. And I where is that gonna go? when you look at how this is gonna done, there’s gonna be a lot of those in the industry that just can’t compete with a hundred twenty-five dollar cost. When you’re doing it, they they can’t compete.
[Pavan] Yeah, that’s right. So, you know, people ask, well, the question I get all the time is, well, if you cost us so low, why aren’t you the number one price? That’s a valid question. And to get there’s more to being the number one price than just cost of production. OK, and a lot of it right now is driven by companies that are willing to just take greater pain and just lose more money so they can get more production. Because, you know, you know how bad the markets are. So we could do more production and make money every month. But the other major component to improving our price is changing our cap structure so that our cost of funds is reduced. Once we do that and a couple of things that we have planned on the execution side, and we’re going to pass that on to the customer. That will make us by far the number one price in the industry. And then it’s basically kind of like for the consumer and for the mortgage broker, this is a no brainer. Fast, simplest service, full spectrum and AI that does everything from 500 credit score, FHA loans to three, $4 million non QM loans all in one conversation. It wouldn’t make sense to be anywhere else.
[David] Well, I could just feel the doubt and unbelief that I’m feeling from our listeners as we listen to $125 to produce a loan operation when everyone else is in the thousands. It it’s hard to believe. But when you said and think about it, listeners, could you have ever imagined a world that we’re in where we’d get in a car that doesn’t have a steering wheel and ask it to drive you from Dallas to St. Louis and it’ll drive you and when it needs more in power, it’ll drive you to a power a charging station and tell you to get out and plug a plug it in. I mean, we’re in a different world. So I think the thing is that the lack of vision. Pavan, your dad pushed you. I loved your dad. I mean that with all the many he pushed me. He was always Dave, I can’t believe you’re coming to me with that argument. That is pathetic. You’re better than that. He always challenged you believing in you. You know you’ve always believed in you. You’re smarter than this. So I mean you look at that and you say that is a wonderful motivator to have that in the background, but you now got this level of success. Where is this going? I mean, where is it possible? Is it I know some more things. We were together in Dallas. We have other countries now interested in your technology and that thing. So this is literally it’s still in its infancy for what it’s going to do for countries and bringing in the part that I know that gets close to your heart and you almost even choke up about talking about is the those that the unbanked population, those that do not have access to credit. And this is where the real power comes in.
[Pavan] Correct. Yeah. So in two weeks, I’ll, mid August, I will be in El Salvador. we have an MOU that we’ve, negotiated. We’ll be meeting. We met last time I was there. I met with president Buccali’s brother and we should be meeting with, we’re setting up meeting with central bank and, president Buccali himself and get the MOU executed. and Angel AI, I will be the banking platform backbone for of Central America and by extension South America because El Salvador is looked up as the leader in technology and governance across Latin America and as we’ve broken through that market as we broke into that market, I believe it would be a domino effect for the rest of Latin America. And we also have a sales team on the ground in the Middle East. And some of those fans got delayed a little bit because of this crazy war. And that has also, as you know, the effect of interest rates and oil prices and inflation and all this kind of stuff. So we pray that this war is over soon and that lives are saved. And we also recently opened an office in Japan and Tokyo. And we’re working closely with some of the biggest families in Japan. They have invested into our fund. And we’re also having conversation with major banks in the Asia Pacific region.
[David] Wow. I mean, where this is going is beginning. It’s gonna open up cap it’s gonna open up credit for those that couldn’t get access to credit. And it’s opening up channels of capital to flow into not only the US, but El Salvador, the Middle East, and many other parts of the country and the best part is these guys can come in and put their capital to work with a confidence that the decisions that are being rendered are warranted. That is amazing.
[Pavan] Correct. That is correct. Banks, is what we call AI you can bank on. And it’s, know, Angel AI is one of them, just like your self-driving Tesla, right? Which Tesla self-driving technology was out 2016, 17, 18, like that. Angel AI was out about the same time, 2018, 2019. It’s a proprietary AI, proprietary neural net that had to be built, that customized, had to be built to handle complex real world scenarios. You just can’t use ChatGPT out the box and plug it in. And you can’t assume that you can throw a lot more GPUs in it and do, you know, run it through more training passes and it’ll get it right. Yeah. and, and,
[David] Yeah. Amazing. Jennifer, you go ahead go ahead. You had another thought. Go ahead.
[Pavan] And anyone who’s curious, like, what really is the difference between Angel AI and its proprietary neural net, its proprietary engine versus the other engines? Okay, just go to angelai.com forward slash comparison. Angelai.com slash comparison. And we have a side-by-side comparison between Angel AI versus, you know, model A and model B. Okay, and these are major, major models, major lenders aand with real scenarios, and you can see the difference. And then you can, beyond that, just go to Angel.ai, go to askangel.ai, and just ask her questions and see for yourself.
[David] It’s wonderful to get to know you have known you all these years, forty plus years, Pavan, having met you when you were eleven years old and now be here with you at this pinnacle of your success. But I keep saying pinnacle, but I you’re not at the pinnacle. You’re still on the mountain climbing up. It’s really Yeah. Jennifer, I have
[Pavan] There’s a lot going on, a lot going on.
[David] Jennifer, You come into this and you are the one that figures it out. You roll up your sleeves, you make it happen. So kudos to you and your team there in Cerritos for all that you do. I love the fact that you guys are all over and growing, continue to grow. So thank you so much for both of you being here. I appreciate the relationship more than you know. It’s so valued. Thank you for the vision. Most of all. Thank you for what you’re doing for average Americans, for those that are getting turned down, that now have a hope and a prayer of getting approved with a warranted product. Thank you, Pavan. Thank you, Jennifer. Bye-bye.
Pavan Agarwal is a renowned leader in the mortgage lending industry and a pioneer in bringing artificial intelligence to the financial markets. Agarwal serves as the President and CEO of Sun West Mortgage Company and Celligence International, LLC.
