In mortgage lending, “close enough” isn’t good enough. While general-purpose AI models like Gemini can generate impressive answers, they often provide probabilistic responses that introduce risk, confusion, and even costly repurchase scenarios. In this episode of the Weekly AI Update, David Lykken sits down with Jennifer, CTO of Angel AI, to unpack the difference between probabilistic and deterministic AI, share real-world borrower case studies, and explain why mortgage professionals must demand precision over possibility when leveraging AI in critical lending decisions.
[David] Listeners, we’re in again with the AI update. We have a very, very special guest today, Jennifer, and if you listen to Pavan. Pavan will be the first to say she’s the brightest person in the room. And I didn’t realize. I just learned that she’s the youngest in her family and she has got a brother that is at MIT and she has an economist in the family, I mean, and they’re all like crazy bright and Jennifer. Pavan says she’s the brightest of them all. So, Jennifer, I knew you were special, but to have Pavan say the thing, especially with his vision and intelligence where he’s at, to have him compliment you like that. It is just such a joy and benefit to have you here and he is so blessed to have you. I’m so excited to have you join the conversation.
[Jennifer] I mean, that is just Pavan’s kindness, right, and his kindness goes to all extent and he says it’s a pleasure to have me. It’s the opposite, it’s an honor to be working with such a visionary.
[David] Yeah, and a visionary he is. I was interviewing Don Healy earlier today and we came up with Pavan and he says and I love Don Healy’s testimony for those of you who have not, you’re going to hear listen to the interview I did with Don Healy of Havonna and it is really exciting, they have, he and Pavan. Donn Healy and Pavon have the same goal and dream of bringing financing and eliminating so many of the middlemen that are taking so much out of the process. But that’s not the topic of this discussion. Today we’re talking about how some people are out there saying that they’re integrating AI, they’re using Gemini, they’re using ChatGPT, they’re using all these things to integrate into their mortgage process, and I talked to Pavan about this. He said no one is better than talking about the flaws in that thinking. The thing that you can use than Jennifer. So, Jennifer, we’re looking to you for the wisdom, as why should people be cautious or leery of relying on Gemini? As good as all of these systems are, as wonderful as they are, as helpful as they are, what is the flaw in using them and relying on them in the mortgage process?
[Jennifer] Sure, I know. Thank you for that great introduction, by the way, David, and as I said, it’s an honor to be here in your presence always. It’s a pleasure to be back here as well. So, when it comes to AI, as Pavan had mentioned in the previous one of the one that you did with him before, there’s a probabilistic and a deterministic AI, right? What is probabilistic versus deterministic and why is it important in the mortgage industry? I’m a consumer. I want to invest in my lifetime savings to buy a property. What do you think is going to make that consumer feel comfortable. Me going and telling them you may probably be able to get this home, you may have to do this, you may have to do that and you may be able to get it home. You know you may have to do this, you may have to do that and you know you may be able to get it right? Is that going to help the customer? or someone telling them you know what, do X Y, Z, and as long as you do X Y Z and able to do X A, B, C, then we’ll do a loan for you. Right, and you stick to this rules, right? That is what is going to help them, right? So this is why.
[David] It’s the certainty. It’s more of the certainty rather than the probably too much of our mortgage lending process, and we have too many out there that are even sounding certain but in reality they’re delivering results that are.
[Jennifer] Probabilistic. Yeah, they’re giving probable maybe messages or confusing the customer. Because, remember, these customers are not mortgage jargons. They don’t understand mortgage jargons. They’re not mortgage experts. Not always right the number of customers who are going to be mortgage experts when they go do their first time buying a home or investing in a property. They’re not mortgage experts. They’re not going to understand everything and anything you’re going to say. So you’ve got to deliver a message that’s simple, that is certain, that’s going to make them feel comfortable about what they need to do. If they don’t leave with that feeling, then there’s no point in investing millions and millions and millions of dollars in AI. That is not going to do that for them, right? So, along those lines, I would like to jump into some scenarios.
[David] Yeah, would you please? I would like to hear that.
[Jennifer] Yeah, sure. So I had two recent situations that in reality came to our tables and I wanted to discuss those two scenarios. I’m going to pick two, right? So one of them was a VA borrower. This VA borrower has two jobs. One job he’s had for a long time and he’s been making money there and the second job he was there initially. He wasn’t on a W-2. He only recently became a W-2 employee with that recent employer, which is like as recent as like January or February, right. And he does get some pension because he was in the military. He was getting some pension income with the military as well. So he came to Angel AI and asked what are your thoughts? How can I qualify for a loan? Will I be able to use both my income? Because I think I need to have both my income in order to qualify for the loan. So will I be able to do it? So Angel AI picked up all the three things he said. One, he has had one job for a long time. Two, he has had the second job, but he only became W-2 very recently. And three, he gets some reserves income. And what Angel AI did was it looked at all the three scenarios and gave a response saying I’ll be able to use your full-time income with the job that you’ve had for a long time the first job and I’ll be able to use your retirement income that you’re getting from VA as a benefit right, but I won’t be able to use your income that you just got switched to W-2 very recently because I don’t have a two years income stability to show. But guess what I can do with that income? Because you’ve received it, for at least you’ve been in that same job and you’re getting it recently. I’ll be able to use that income to offset any debts that you may have which are anywhere between 12, 6 to 24 months. So that is the key factor there, which is she was able to look at all three information that the consumer gave and gave specific responses about what she can use and what she cannot use and what you could do with the income that she was he was not going to be able to use as a qualifying income, right? so someone who’s getting such a deterministic answer or such a like a complete answer, makes that consumer feel much comfortable and go and do the loan, go forward with the next step of writing a contract or, you know, applying for the loan or mortgage or whatever it is. And the second scenario that I wanted to talk about is I had a consumer who come to me and tell me that recently they have they. This person has been in sales for a very, very long, long time, since 2017. It’s about eight, nine years. The issue is originally they were making bonus income right, and then, for a period of time within the last two years, they were not making bonus, they were only making commissions. Now, in the most recent seven months, they’ve started making the bonus income again. They actually had commission income just for the last seven months, but they’ve had history of having commission income before two years ago, right. So from 2017 to currently they are in sales. From 2017 to somewhere around 23, 24, they were actually making commission, but from 2024 to 2025, in the last two years they’ve only had bonus income and they switched to another job where they started making commission income since February of 25. So for the last eight months they’ve made commission income. But if you look at the two-year history, I don’t have a two-year history of commission income, but I’ve had commission income before the two-year history, so I’ve been in the job for eight years. So can I use the commission income is a key question here. Angel AI was able to quickly pick up on the fact that, even though this customer has had an experience in the sales and they’ve had commission income before, they have not had commission income within the last 12 months continuously. So she was able to quickly tell them hey, only because you’ve had this commission income for less than 12 months. So she was able to quickly tell them hey, only because you’ve had this commission income for less than 12 months, I won’t be able to use that, but had you had the commission income for 12 to 24 months, then I could have been using that because you’ve had income before, in the current scenario you won’t be able to use the commission income because I have to document to your history, but I could have used 12 to 24 months had you had it at least for 12 months. She was able to pick up on these two things. Out of curiosity because every lender out there is very proud of their relationships with big companies like Google and they’ve signed up with all these lenders and all these AI companies and they’re very proud of those things. I just went and tried to do a Google search of exact two questions Right and I’ll share my screen to show you what I found Right. On the first question for the VA borrower, I’ll share my screen to show you what I found right. On the first question for the VA borrower, Gemini came back with some very convoluted response. And if I put my With you, being a mortgage professional I’m lost. I was like okay, what are you talking? Can I use it or not? I simply need to know whether I can use it or not. So Gemini didn’t give me an answer, right, didn’t give me any suggestions. It gave me broad guidelines and whatnot. I was like I don’t need the guidelines, I’m a consumer, I don’t need guidelines, right? And the second question is even worse.
[David] They gave him an approval.
[Jennifer] They gave me an incorrect response. Imagine an underwriter who is relying on that and use that commission income right. One of two things could happen. One I will have a borrower who signed a contract, gave the EMD, and guess what? Someone woke up all of a sudden at the 11th hour and said, oops, I can’t do it. And the borrower’s EMD has gone.
[David] How many times have we seen that play out in this industry?
[Jennifer] Exactly, or even worse, the loan closes because the underwriter relied on the response that was given by Gemini and went ahead and closed the loan. And guess what?
[David] The loan’s now not saleable, yep.
[Jennifer] Either not saleable or a repurchase is on the way. Imagine the number of loans that they’re going to get stuck with just because they’re relying on an engine.
[David] You’re really bringing up a really good point, Jennifer, is that you know as good as these tools are for certain aspects of helping us construct different things, this will not work like what we have come to understand, respect and value and trust in Angel AI, this is a great. Those are great. Two examples, Jennifer wow.
[Jennifer] So I want to show you the facts, right? I don’t want to just keep saying so if you don’t mind me sharing, no, please do. I’m going to share my screen, okay.
[David] Oh, there you go, wonderful, wonderful. I wanted to show something.
[Jennifer] So these are the two scenarios that I tried with Angel AI and I was able to get the questions that we’re talking about right. So this is the first question where I’ve had a commission Can I use the commission income? Right? So I went and asked Angel AI the question and you can see clearly the response that she has given here, right, and please let me know if you want me to blow up the screen so you can see it clearly. But that is what she gave you right. And then I went and asked her the VA borrower question as well and see what she gave me. She clearly said that one two, first and second, I can use. The reserves can be used, but I can’t use this. I can only use it to offset any debts that are within six to 24 months, and the same thing here, with a six month of commission history. She was very certain and told them that I cannot use it from the current job, but I could have used it if you had, at least for 12 months because I can use the compensating factor Very simple, clear. Now let’s look at what Jim and I gave me, what Jim and I gave you. So this is on the question for the VA borrower, right? I pasted the exact same question here.
[David] Same question.
[Jennifer] Like if the complex income structure approval may be possible by presenting a strong case of stability based on career field continuity, reliable income, a borrower is not going to understand any of this. I mean, it’s way too much information for a borrower. All they need is simple, clear direction, which I wasn’t able to get here. And two, if you look at this scenario where I talked about the commission income based on the borrower’s employment, it’s kind of conflicting here. If you see, it. It is possible for the commission income to be used right, even though they’ve not been at their current job for less than 12 months. But guess what? Then they have the guideline here conflicting information. If I were to make a decision based on what Google just said or Gemini just said, I’m screwed.
[David] Yeah. That is such a good example, Jen. Listeners take the time to go to a computer screen and watch not only listen to this, but go back and watch this, because it’s such a good example of two scenarios one that could get done and one that couldn’t get done. It shouldn’t have gotten done. And, Jennifer, I’m just so grateful to have you here and sharing this.
[Jennifer] Just like to add a couple of two points, if you don’t mind. So what I would like to add to that is this didn’t happen overnight, right? Angel AI didn’t happen overnight, It involves a ton of research and a lot of knowledge that SunWest, as a lender, had gained. It’s all based on that 45 years of history of knowledge and an amazing group of engineers. I mean, if you recently read the news where Mark Zuckerberg paid like $250 million package for a research engineer, right, this is why angel AI is valued at tens and billions of dollars, right? So it’s not easy. So, building AI that is going to be able to give you an answer, that is going to make the customer feel confident and, at the same time, you, as a lender, is not going to get stuck with a bunch of repurchases, can’t just happen overnight. You can throw any amount of money at it, but you can’t make it happen overnight. You can be running GPUs at millions of dollars. None of that is going to matter when your underlying engine is flawed. This is why, if there are underwriters out there or anybody out there, Angel AI is free to use. Go ahead, try it for yourself. Use it and make yourselves better, because we want to make everybody better. We’re not here to take anyone’s job
[David] I don’t know if Pavan shared with you, Jennifer, but I was at a dinner party in reno, here celebrating the birthday of one of my clients and he had brought his, his opts team around him and the topic of Angel AI came up without even me prompting it. And someone said have you guys heard about Angel AI? And this gal goes. I use it all the time and I was afraid to tell anybody and I said oh, I know Pavan, he’s gonna love that. Oh, please don’t. She was just terrified that they’re gonna get in trouble. I said, no, that’s why Jennifer and Pavan created. They want it out there to be used. And listeners, I’m serious, take this, get signed up. You can start using it for free, going to angel.ai, and you can actually start using the technology. Use it for scenarios, Pavon, Jennifer, the whole team wants you to go in and continue to put loan scenarios in there. And people say, well, aren’t they going to steal the loans from me if I do that? No, these people are with the highest degree of integrity, I’ve ever seen of anyone in the mortgage industry, and that’s a bold statement. These two, I tell you, the whole company is designed like that and it comes from Hari, Pavan’s dad. He instilled this in them and it’s so exciting to have you, Jennifer. You’re the CTO, you’re the braids of the power behind it, and we’re so thrilled to have you here. Would you start coming back regularly on these AI updates so we can hear from you more regularly?
[Jennifer] Sure, it’ll be my pleasure, David, and being able to share what Angel AI is for and all about her is my pleasure. She’s a baby.
[David] It’s an honor to know you. It’s an honor to work with you all. I’m so excited about this. People be aware out there, you can use Angel for free and it’s proven, tested over 20 years. Forty years of mortgage lending has gone into it. Twenty years of development, millions I think over 100 million now has gone into the development of it. It is so important that you use a tool that you can trust. Don’t trust the other LLMs out there, the other open, the Gemini’s and the open AIs to make important critical decisions, and Jennifer just gave us a good example of that. Jennifer, thank you so much, appreciate you.
[Jennifer] Thank you, it’s my pleasure. It’s my pleasure.
Important Links
- Press release about Japanese investor bringing AI powered innovation
- SunWest Mortgage website
- Angel AI website
- LinkedIn – Pavan Agarwal
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.