- 7 hours ago
Artificial intelligence has quickly become the most talked-about technology in mortgage lending, but there are questions on where it actually delivers value. In this HousingWire conversation, Nima Ghamsari, Founder and Head of Blend sits down with HousingWire’s Allison LaForgia to unpack where AI is truly reshaping the mortgage origination process and where adoption still has a long way to go.
Ghamsari explains how AI agents can move beyond simple automation to handle complex workflows across the lending lifecycle. The conversation also explores Blend’s newly launched Autopilot, what it does differently from traditional rules-based automation, how lenders implement it, and why Ghamsari believes this is just the beginning of a larger shift in how mortgage operations will run in the years ahead.
Ghamsari explains how AI agents can move beyond simple automation to handle complex workflows across the lending lifecycle. The conversation also explores Blend’s newly launched Autopilot, what it does differently from traditional rules-based automation, how lenders implement it, and why Ghamsari believes this is just the beginning of a larger shift in how mortgage operations will run in the years ahead.
Category
🗞
NewsTranscript
00:06I'm Allison LaForgia from Las Vegas, Nevada, and I am sitting with Nima Gamsari, the founder
00:13and head of Lend. Nima, thank you so much for joining me today. Thank you for having me. So
00:17over the last year, everybody has been talking about AI. It's all over the industry. Where do
00:24you see AI reshaping, origination, and where do you think lenders are right now with adoption?
00:32Let's start with the second question first. I think AI has evolved so quickly, even in the last
00:38four months, that most lenders, and we work with a couple hundred lenders across the country,
00:44are just not really there yet. And it's not there. I mean, the industry is shaped so quickly,
00:50but it's also a golden opportunity because they do spend hundreds or thousands of dollars per
00:54file reviewing documents and doing stare and compare and all those things. But I don't
01:00like to, when I talk to our lenders, our customers about this, I don't like to talk about becoming
01:04AI native. I like to talk about becoming agent first. And the distinction of being agent first
01:10to me is instead of humans taking the first pass of the stare and compare and then running
01:15it by an agent or jumping in their computer to send it to an agent, we want the agents to
01:21be listening and watching behind the scenes and doing a first pass before a human even touches it.
01:26So by the time it gets to a human, it's already really ticked and tied. And so that's been a
01:31big
01:31focus for us, making our customers agent first, because they do spend too much time and too much
01:36money reading documents, requesting documents, calculating income, finding information, and it's
01:42just a waste of their energy. Now, there is a lot of debate over where AI fits in the mortgage
01:48tax
01:49stack. Where do you think AI agents create the most leverage and why? I'd say our focus right now
01:58at Blend with Autopilot is to make it so that the underwriters and processors and the consumer
02:05are 100x more efficient. And what that means for the consumer is the first time they show up,
02:10as they're going through the process, it's dynamically understanding what they need.
02:14And then as they provide those things, if new things are discovered, providing that to the
02:19consumer and understanding the file in real time so that they're able to get through it in one 30
02:24minute setting instead of five 20 minute settings. So that's for the consumer. For the processors and
02:29underwriters, same thing. By the time that they get a file, we want everything the consumer has done
02:34to be fully underwritten, show up in a nice report for them so they can say, yeah, this, you know,
02:38Nima's good on his mortgage and we already have his income verified, his assets, his credit.
02:42And so by the time they get it, it's a 20, 30 minute exercise as opposed to a multi-hour
02:47exercise
02:47with four or five iterations they have to do when they find new things. Instead of going back to the
02:52customer, it's already been done by the AI agents. Nima, you just mentioned Blend's launch of Autopilot.
02:58Congratulations. But I want to dig in a little bit more about what it actually is and what it does.
03:04Sure. So it's really simple. Like I said, I want to have our customers be agent first. And so it's
03:12ambient
03:12agents, agents that are working in the background as the consumer's entering any data field, as they're
03:18uploading any document, Autopilot's reading it, understanding it, underwriting against the guidelines
03:23and other fraud and other kinds of things that have to be checked, regulatory things. And in the end, what
03:29that
03:29means is the consumers get a perfectly tailored, hyper-personalized needs list and experience for
03:36their specific situation. And what the underwriter gets is a fully underwritten file with a nice summary
03:42of all the work the agent did while they were sleeping. It's super simple. It's working in the
03:46background to make their lives a hundred times more efficient.
03:49I want to dig in a little bit more with Autopilot. And can you walk me through how that's different
03:55from rules-based automation? Sure. Yeah. So we've been building Blend for 12 years now.
04:02And our first take towards this problem 12 years ago or 10 years ago was something called
04:08Blend Intelligence, which is exactly what you described as rules-based automation. We would
04:12look for inquiries in the credit report. We would look for large deposits in the bank account.
04:16And we'd be able to do it if we had data. But then we didn't handle any of the two
04:21big
04:21distinctions. We couldn't do it if there was no data. There's too many kinds of bank statements.
04:26There's too many kinds of pay stubs. So if there's no data, we couldn't really apply it. That was the
04:30first thing. And the second thing is that as the rules were being written, we only were able to write
04:36them for the majority cases and not the edge cases. And if you ask any underwriter, they say they spend
04:41all of their time on the edge cases. And that's where Autopilot shifts that on its head and says,
04:46we don't care what kind of document it is. We don't care what kind of data it is. We don't
04:49care
04:49how complex the person's income is. We're going to solve this for you. And it's going to show up on
04:55your desk perfectly cleanly. And just as an example, I ran myself through the normal Blend
05:02process, which is what Autopilot is listening to. And I have pretty complex income. I have multiple
05:08K-1s. I have multiple 1099s. I have a small business. And obviously, I get RSUs and things with
05:15Blend shares with Blend. And it was able to perfectly calculate. I ran it a hundred times
05:21to say, is it different? It's another important point. But you don't want us to have different
05:25answers from AI in different times. You don't want AI to hallucinate. I ran it through a hundred times.
05:29I got the same exact answer from the AI every single time. And it was the exact right answer
05:33against the exact guidelines that are required. And so for me, that would be a great experience because
05:39I'd know exactly what I'm approved for. And I'd know what the underwriter is going to find right
05:44up front. And for the lender, it's great because reviewing my K-1s and 1099s and my small business
05:51tax returns and all those things was going to be a bearer of an effort for them.
05:56Now, this sounds like it could be a lot to get up and running for a lender. What does implementation
06:01actually look like? Well, the nice thing is all of the consumer side documents at least
06:09are provided to the lender by the consumer via an upload mechanism or something in Blend already.
06:15So if they're already going through the process and they're applying for the mortgage
06:20as they normally would in Blend, and then they're uploading their documents, they get their needs list,
06:25and it has the W-2s and the tax returns and pay stubs and whatever other things that the system
06:29decides or autopilot decides, the lender doesn't have to do anything. It just works behind the
06:34scenes. They flip a switch to turn on. They have to obviously want to turn it on. But once they
06:39want
06:39to turn it on, it's truly a flip of a switch for them. And this is one of the first
06:43products that
06:45we've launched in that way, if not the first product. And we wanted to make it really easy to
06:52adopt, really easy to try. We're giving it away for three months during this preview period just to
06:58let people try it because it's so hard to understand, like, how is this even possible?
07:02Even for me. Until you have it running. Until you have it running. And then you see yourself going
07:06through the process or you test it with a real borrower situation that you've run into that's
07:09cost you a bunch of time and energy. And you're like, wow, this, you know, this AI stuff, these
07:14agents actually do work. So it is, it is, we wanted to make it as easy as possible because
07:20there's too many other things going on in the world. There's like a war going on. There's all sorts of
07:24things going on that, you know, people want to make things, I want to make things as easy as
07:28possible for our customers. So you just mentioned how it can be a little hard to understand how
07:34you're going to use it unless you're already in it. Talk to me a little bit about the reactions that
07:38you've already seen from lenders and customers who've already seen autopilot in action. Yeah, sure.
07:46Actually, I'd say it started with our customer advisory board, which we do, you know, twice a year.
07:51Um, and we had it about a month ago, mid Feb. And, um, we had our customers there and they
07:58were
07:59asking about it. We were demoing it and they were asking about it and their eyes kind of lit up
08:02and
08:03they were like, wait, wait, so this, this will fully underwrite the file without my people having
08:07to touch it. And it'll show up on my people's desk with a, you know, something they can react to.
08:11I'm like, yeah, that's, that's how it works. Um, and then in early March, we made it available and
08:19customers, people were skeptical that lenders would just turn this on. We had seven large
08:23customers turn this on in the first week without even calling us. Just turn it on in their
08:27environment. Try it. We went live in production. Very recently, they turned it on in production.
08:33You know, it's like when you make it easy for people to do things, they'll do it. And I think
08:38that's, that's kind of the tricky thing around AI right now is that there's a lot of so much hype
08:42around it. So there's so much, and that also has a, not a leads to a lot of noise. Yeah.
08:46To your
08:46point. And so it's just difficult to make ends like to make sense of what needs to, how to actually
08:53take advantage of it. And so anyways, my approach was let's make this as easy as possible for our
08:58customers. And we're seeing them both with their feet. They turn it on and they're like, wow, this
09:01is amazing. And we're paying, you know, tons of money underwriting these files. And my team can only do
09:05six files a week and they could probably do 50 files a week with this kind of technology.
09:09So autopilot sounds amazing. I'm going to ask you for a little bit of a teaser, a little bit of
09:14a preview. Where does this go from here? Is autopilot the beginning of something bigger for
09:19Blunt? Yeah. I, you know, I, I think getting agents involved in the, the mortgage lending process
09:26is good for definitely good for underwriting and processing, which is kind of where we've started
09:30helping the underwriters and processors get their work done behind the scenes while they're sleeping
09:35on making them hopefully a hundred times more efficient. But I think this should apply
09:39to all parts of the loan process. So as a consumer is going through, if they switch loan products
09:45and that leads to additional things coming up, like they should know that in real time,
09:49Hey, if you're going to go and you're going to lower your down payment, we're going to have
09:52to add MI, but something should be telling them that before, you know, at 2 AM in the morning
09:56while they're trying to work through this file. And so one of the things that's exciting
10:00to me is how do we take this technology and make it so that every single consumer has a very
10:06personalized, very tailored experience that's generated on the fly for them. And so this
10:12concept of generative UI has been around recently in the last six months or so in, in the AI world.
10:18And it's super powerful because your financial situation is different than mine and your questions,
10:22the things that scare you, I'm a first time home buyer. Maybe you're not, you've been through
10:26the process a few times. Maybe your, your process is a little bit less guided because you've already
10:30been through it and you're like, Hey, I just want to walk, just tell me what I need and provide
10:33it.
10:33But for me, I want to be walked through, Hey, if you had $2,000 more in down payment,
10:37that would lower your rate a little bit more because that gets you into a lower rate band for
10:40us. Um, and so those kinds of things can be dynamically generated on the fly. Now it wasn't
10:45possible. That's where humans actually were doing a great job. And now the same humans can be,
10:49you know, 10 or a hundred times more efficient. That's amazing. Something that we talk a lot
10:55about at housing wire is how each borrower is really different. Even if you have different
11:00institutional investors, somebody buying their first time home, somebody buying their second
11:04home, wherever the transaction is, everybody's a different profile. So having a technology that
11:09assists you in meeting them where they're at is so critical in an age where we have such high
11:15expectations for customer service. Yeah. I couldn't agree more. And, and by the way,
11:20it's, it's critical for the customer experience, but it's also critical, you know, for a lender,
11:24I think because one, you want to give them that level of service, but that's a customer who wants
11:29to do business with you and you might lose them. They might be on the hook and you might lose
11:33them
11:33because of something you didn't tell them that someone else tells them. And so I think being able
11:37to provide that really tailored experience is everyone's dream. It's just that doing, you know,
11:43millions of phone conversations or meetings with consumers doesn't scale. Absolutely. And, you know,
11:49it doesn't mean you don't want to have a human touch. I think humans who want to have a human
11:52touch
11:52should always have that ability to have a human touch and such a big life process,
11:56but we shouldn't force people down the path of, well, if you want to work with us,
12:00you have to schedule a time in two weeks, right? Like that's, I think that, that world is sort
12:04of going away and people are self-serving, whether it's dog veterinary care or whatever it is
12:09on chat CBT, they don't just call their vet first. You know, it's like, they're just finding
12:12ways to help themselves where they can. Well, Nima, thank you so much for sitting with me today.
12:17Congratulations on the launch of autopilot. And I can't wait to see what's next for blend.
12:21Yeah. Thank you. Thanks for having me.
Comments