Mortgage lenders can’t wait on AI. But they shouldn’t trust it blindly

Name: Mortgage lenders can’t wait on AI. But they shouldn’t trust it blindly
Uploaded: 2026-09-25T08:00:34+00:00
Duration: 13 min 56 s
Channel: HousingWire

00:09I'm Allison LaForge with HousingWire and in this episode of 10 Minute Talks, I'm sitting down with

00:15Julian Hebron, the founder of The Basis Point. Julian is partnering with HousingWire for the

00:21Mortgage Banking Summit on October 1st. We have some really exciting demos lined up that we will

00:27be doing live from onstage, which is why I brought him on 10 Minute Talks to really dig into what

00:34we

00:34should be paying attention to when we look at technology as an industry. Julian, thank you for

00:40joining me today. Allison, I always love joining you, so look forward to chopping it up a little

00:46bit today. Me too. I always love a moment where I get to pick your brain and I am excited

00:55to talk to

00:55you specifically because we know that we're at a point where mortgage companies have invested

01:01heavily in technology, but origination costs, as we know, are still incredibly high. What do you think

01:09the industry is getting wrong? Yeah, and I will, I'm going to hit that and I'll just do a quick

01:14little

01:14teaser of Housing Stack Live too, because Allison, I just want to echo what you just said for anybody

01:22listening, watching. October 1st, your Mortgage Banking Summit, the HousingWire Mortgage Banking

01:29Summit does feature this program called Housing Stack Live that the Basis Point and HousingWire

01:33created together, which does a very rapid fire state of mortgage technology and AI, as well as live

01:40demos. And it's always fun to do this stuff live on stage because when I'm hosting these companies on a

01:48big stage like yours, we can both advocate for them, but also keep them honest on stage. And as

01:55we go through some of my answers today, I'm going to, I'm going to give some of those honest answers

01:59as a setup for all, especially the fintechs listening, because if we're not honest to the

02:05lenders buying all this stuff, um, the, if the basis point is being too big of an advocate for the

02:10fintechs, then it doesn't help anybody. So that's just a little context of how it plays out,

02:15how Housing Stack Live plays out on your stage. Um, and it's going to play out today in this 10

02:20minute talk, but yes, going back to your question in terms of heavy investment in technology, um,

02:28frankly, really since, um, going back 10 years really, um, and the AI era has only accelerated that.

02:36And here's your, here's some answers to your questions on the origination costs, right?

02:41So I went back and looked back to 2008 and cost per loan was $4,884. If we adjust that

02:50for inflation,

02:51it's $7,467 today, cost per loan is 10,936, right? As of 2Q, that's the latest data available

03:00from MBA performance report. That is 46% higher than say 2008. Um, what has been going on in all

03:09of

03:09these years is the addition of technology and Allison, here's the core answer to your question.

03:14Tech's only 4% of that tech is only 4% of the $10,936 per loan that MBA reports

03:23every quarter.

03:24Okay. I think people forget that. So what are the levers for cost control with lenders? Sales

03:31fulfillment, fulfillment, production support is 61% of that $10,936. Sales is 30%. Fulfillment is 13%.

03:43Production support is 7%. Managers and directors are 4%. Okay. I think to directly answer your

03:53question sales. If it's a retail shop, those costs, I don't, I don't think are going to go down.

04:01Everyone is always talking about, Oh, how are we going to pay people less? If they're bringing in

04:05the loans, you're not going to pay them less. It's just that simple indirect maybe, but you're just

04:09not, uh, in retail though, fulfillment, production support and managers. I think in the AI era,

04:16the smart fintechs and smart lending executive teams, this is what they're focusing on. I don't

04:23mean to be heartless about the jobs, but everyone knows that AI impacts certain jobs. And I think

04:28those jobs are the jobs that can be more, um, impacted would be more impacted if technology

04:35can get it right. And we haven't so far. That's why those costs are what they are. But I think

04:41it

04:41comes from fulfillment, production support and managers. This is where the AI efficiency will trim

04:46those costs. And that's where you get there. Now, Julian, we know that there is no shortage

04:52of new technology entering the mortgage ecosystem. How do you separate the technologies that are

04:58actually moving the needle from the ones that are simply adding another layer onto an already

05:04complicated tech stack? Yep. Uh, things definitely, definitely feel crowded to the basis point, especially

05:11in, um, what we call like loan production optimization, which is a number of things.

05:16There's just lots of teams chasing the same things. And to be blunt, they have been for a long time

05:21before the AI era. It was the same thing. The AI era has brought a whole new crop of teams,

05:27um, that, that are ending up kind of going after the same problems. And the good news is that you

05:33do

05:33need that. Like you need AI native POS modernization as an example is like a key one, right? You do

05:41need

05:42that. So yes, it's crowded, but yes, there will be the same kind of thing that there has, has been

05:49in

05:49the previous FinTech cycles where you get a bunch of teams going after similar problems. And then there

05:54is some thinning out as certain teams become category leaders. To me, the combining of POS, CRM,

06:03AUS, and verification of income, employment, and assets, that's four cat, four big categories that

06:10are all, Oh, they're all, they all overlap each other. And so the, the, and, and the lenders would

06:16love to have, you know, to go from four to three and three to two. I don't think you're going

06:21to go

06:22down to one category that covers all of that, but between POS, CRM, AUS, and BOIEA, I just think that

06:29that stuff has to thin out for lenders to get real efficiency. Um, but of course it's just, it's way,

06:35way harder than it looks. I won't name names today on a 10 minute talk, but if anyone ever wants

06:41to

06:41reach out or Alison, if you and I want to do a follow-up and we actually do a 10

06:45minute talk where I

06:46start naming names, I can, I can talk about what teams that the basis point thinks are making real

06:53progress on combining of these core categories for loan production optimization. Sounds like we have

07:00a part two that we're going to have to figure out. Yes, for sure. So because it is its own

07:06topic,

07:07but that's the setup. It is. And it's, it's an important one to get into. And it's interesting

07:14because it's a really core part of my next question for you, which is really getting into

07:22the lenders who are getting the most value from technology. What are they doing differently,

07:28particularly around implementation and adoption? Is there a point where being too cautious about

07:35adopting new technology becomes its own competitive risk? And do you, how do you think lenders should

07:41think about when to move versus when they should wait? I think that's kind of a two-parter. And I

07:48want to answer the second part first, because is there a point where being too cautious about,

07:54and I'll just change your wording a little bit, adopting new AI, because this is the thing AI has,

08:00it's blisteringly fast, but it's, but it's, it's a, it's the wild West to say the very least,

08:07right? Because it's just, it enables anybody to do anything. So as for, you know, waiting or being

08:14too cautious, my advice to all lenders is do not wait. But do not think that you know what your

08:22LOs

08:22want anymore. This is the most important part. Like you can't wait, but you, your LOs are using AI on

08:31their own with or without your consent. They are, and they're figuring stuff out. And that part is

08:37where you need to let this stuff bubble up. You need to let your sales teams come to you and

08:43tell

08:44you what they're using and not make it feel like they have to do it in secret. Because if you

08:49do,

08:50you are not going to let the best ideas bubble up. And so what's happening is that AI is making

08:56everyone that, that, so that's, that's for the exec side, if you will, that's the exec side answer

09:02is let it play. Let the AI cowboy stuff play. You have to, because if you do, you will let

09:10the good

09:10ideas bubble up and you will get the best things that your LOs and your, and your sales and production

09:18and, and frankly, fulfillment teams overall really, really want. Okay. My advice to LOs though,

09:24is this, you are using, you are using this, your own LLMs every day. AI makes everybody an engineer.

09:34And this is pretty exciting. But LOs don't let AI make you think you're an engineer. You're not an

09:42engineer. And I will just remind everybody about, think about how AI answers your questions. You're

09:49always right. And it's always very exciting when you get the answers from AI telling you how right

09:55you are. But I have noticed in the work that we do at the basis point that AI is very

10:00often

10:02confidently wrong. I'm going to say that again, AI is too often confidently wrong. So you're not an

10:08engineer, you're a salesperson bringing the loans, but use AI and tell your teams about it because what

10:15you are doing, especially the smart teams, they are vibe coding, cool stuff, get that to your

10:20mothership and let them operationalize your ideas at enterprise grade. That's how the game should be

10:26played. Now. Speaking of how the game should be played now, Julian, if you were sitting in a lender's

10:33executive suite today, and you had to prioritize three technology investments for the next 12 to 24

10:40months, what would you be looking for? And what questions would you ask before spending the money?

10:49Okay. I'm going to start with what, like if I had, had, had, had to choose one, just one, not

10:54three.

10:56It would be, it would be probably simpler than you might think. It's how to make LOs better at AEO

11:03and

11:03GEO, which is like answer engine optimization or generative engine optimization. This builds on all the

11:09basics of SEO, but it's more because AI is, is, is gathering everything it knows, say about a loan

11:17officer or a company. And it starts with things like reviews. And then it goes to what's the

11:22reputation of the LO or the company. And then it goes to the geography, where are they? And then it

11:27goes to content. SEO is always like, you know, this idea of like, put out a bunch of content and

11:33yes,

11:34all of that matters, but that's kind of lower on the list about how AI bubbles up in, in AEO

11:41and GEO

11:41methodology. If you want people to find you, it's all those other things first. Right. And so to me,

11:48that's the one investment either that you need to make at an enterprise level or enable your LOs

11:53to do it, or, or there again, Alison, there are tools that can do this. And if we wanted to

11:58do a part

11:59two, we can get into some of the category leading tools that are enabling this in, in smart and to

12:05your key question, affordable, like before I spend the money sounds cool. How much is it? And is it,

12:10does it really work? There are tools that do this, right? So that would be my one. And then there

12:15would

12:15be kind of maybe two others. I won't even go into the weeds on them, but once you get good

12:20at that,

12:20you're going to have more kind of cold leads, right? So speed to lead and conversion would be number

12:25two for me. And, uh, there are some systems that are doing this well, and we've needed this for a

12:32while, but how do you not sound like AI when you're responding to colder leads where it takes longer,

12:39you have to drip on them a handful of times before you're going to close them. There are some smart

12:44things going on in that area right now. And then the last one is file quality before submission. I

12:49touched on it in production optimization. When you asked me that part, um, I'm going to close

12:55on that part by saying, I think it's super important, but if my previous answer didn't

12:59tease this, I'll say it explicitly now, which is like, I think you still need people for file

13:04quality before submission. I still do. I think you need killer processors that no credit. I just do.

13:10And, uh, that's my one old school view for as much as I like to support innovation, both with the

13:15basis point and all the work that we do with housing wire, but that's my one old school take.

13:20So, well, Julian, thank you so much for joining me for part one of this conversation. We're going

13:26to have to bring you back to get more into the weeds and thank you for setting the stage for

13:31us

13:31about how we should look at technology and how we should look at some of our strategic investments

13:37as we head into the mortgage banking summit on October 1st.

13:42Absolutely. And I can't wait to do housing stack live on your stage and support the fintech community

13:47like we do. Can't wait to have you and see you then.

Mortgage lenders can’t wait on AI. But they shouldn’t trust it blindly

Category

Transcript

Recommended