Maximizing Returns on AI Investments With Seever Sulaiman | August 2025 Data

Olivia DeLancey (: 00:22

Welcome to the Market Advantage podcast. This month we're going to look at data highlights from August. Then we're going to sit down with our guest. have Optimal Blue's Chief Technology Officer, Sieber Silliman, us. My friend and welcome. Great to see you both.

Brennan O'Connell (: 00:39

Likewise, Olivia, as always.

Mike Vough (: 00:41

Excited to be here.

Olivia DeLancey (: 00:43

So we have the Labor Day holiday behind us. What did we see in August in terms of data?

Brennan O'Connell (: 00:48

Well, we had sort currents moving in different directions that netted out to a 2 % decline in total rate lock volume. We had purchase activity sort of trailing off here as we passed the peak buying season, we're down 10 % in August relative to July. And that outweighed a pretty strong improvement on the rate in term refi volume side. So we were up 70%.

with rate term refinances, given the improvement in mortgage rates. rates, a little bit of rally. know there's a lot of discussion around upcoming rate cuts. Market seems to be building in some expectations around those. The OB-MMI 30-year conforming rate.

That's the benchmark for the CME OB 30 mortgage rate future. Finished the month right about six and a half down nearly a quarter of a point from July. A few other rates here, Jumbo rates dropped 30 bps to 6.57. FHA fell roughly 25 bps to six and a quarter and VA was down 33 bps all the way to six flat.

Hopefully we continue the rally on the right side and can start to sniff some five handles here in September. A few other odds and ends in terms of production. Conforming gave up a 1 % share.

of its total market production, primarily to VA and then a little bit the non-agency space as well. then PUD volume, we've talked about this in the past, but the trend continued, the decline here, we're seeing less and less planned unit development in our rate lock volume.

That's a really good proxy for new construction. So the mortgage lending that's related to new construction continues to fall. It's from a market share perspective, it's down to 28%.

It was up near a third last year. So at the same time last year, so we're, we're down from roughly a third of production to starting to creep towards a quarter of production, but being related to new construction. So I do think that that's probably telling us something about what's going on in the builder community.

availability of some of the homes, the lock-in effect may be starting to loosen a little bit. So more of the inventory that's available for sale starting to come from existing homes versus new construction. Otherwise, a couple other items, first time home buyer percentages were mostly flat. VA took down a bit. And I just say the non-QM production, slow and steady increase here. We hit 8 % in terms of total.

production from non QM last month. It's up to about eight and a half here. So I don't think that's a trend that's going to be changing anytime soon. So for folks that are not yet in the non QM game, I think, I think everybody will be in some shape or form here in the near term. Mike, that's it on the origination side. So I'll pass it over to you for the secondary stats.

Mike Vough (: 03:55

Yeah, thanks for passing that over to me, Brian. think the one thing on non-QMAT interesting is just to kind of take a step back and think about where it started. You go back to like the COVID era time, that was like a one and a half, 2 % of originations. And just to like see that, how much that's grown in the last, you know, three or four years, it's ⁓ definitely something that seems like it has some staying power. But over to the secondary side, know, a big number that we saw change this month was in

And we continue to see a shift to MBS securitization. It moved from 37 % to 40 % of the loan sales that we've observed in our hedging platforms. And it just kind of continues to beat home that point where we might be seeing some market shift from, small to mid IMB, small credit unions and banks to maybe more of these larger institutions that depositories that have that allow them to shift.

you know, the rates that they're able to give folks, or you're seeing the very large, IMBs, publicly traded companies, really starting to shift their economies of scale, shift that muscle that they have there. saw cash of sales drop 2%, then we saw bulk loan sales to other aggregators drop 2 % as well. oddly enough, we also saw best efforts increase about a percent, which is interesting as well.

stat to think about too, that I want to call out is the shift in loan sale by price. So this month we actually saw quite a large increase in the loan sold to the highest price that went from 70%. That was at 70 % last month and now increased to 75%. So that softer eligibility things like representative mix, trying to make sure all your investors get equal slices of your origination markup.

is starting to see a decrease where it seems like lenders are chasing price and they need to chase as many basis points as humanly possible in terms of maximizing their gain on sale. One more stat that I wanted to call out was what happened with servicing rights. So mortgage servicing rights move in line with rates. as mortgage rates drop, the value of servicing

will drop with it. with OBMMI dropping quite a bit this month, we saw on average the servicing rates recognized at the point of sale drop almost four basis points this month, a shade under five basis points, went from about 1.2 to about 1.15 a percentage of price for servicing rates. On average, OBMMI was down somewhere between 14 and 15 basis points this month. When you look over the course of the month, we know it ended about a quarter of a point down.

over that period of time, we saw the servicing rates drop about five basis points in value. And that's coming out of people's margin. The value of servicing is something is a lever that folks can pull from a profitability standpoint in their margin. An interesting thing also on this point is we saw the primary secondary spread widen this month. So what that means is lenders as rates drop, they're unsure how

how long they're going to drop and they're unsure if they have the capacity to handle more origination. So they'll widen that profit margin. So what that means is that even though it dropped five basis points, the average servicing value, we may see it drop more as that primary secondary spread tightens if we're in this rate environment prolonged period of time, which would be an interesting thing to kind of track there.

Olivia DeLancey (: 07:15

Great, thanks guys. So I think it would be a good moment to talk stats of the month. Anything stand out to either one of you that you want to share?

Mike Vough (: 07:23

An interesting stat that I wanted to call out was the share of loans eligible for a spec pay up. This is not something that we are tracking just yet in the market advantage. this our trial balloon. But the agencies recently came out with characteristics eligible for a spec pay up. back in the day, it used to be if your loan was less than 200K, you would receive a pay up.

And the idea there is that it takes a lot of rate move for those loans with low loan amounts to actually refi because the payment savings typically don't outweigh the closing costs. You're not going to pay 10, 15 K of closing costs if you're only saving 50 bucks a month in payment. It just takes too long to make that investment work. But as we've seen home price appreciation and inflation go up, we've seen investors and buyers of mortgage backed.

securities go higher and higher looking for that call protection, they call it. And recently Fannie and Freddie both released a 325 and a 350k spec pay up. And so the average loan amount that we see in OB is somewhere in that 380 percentage. So one of the things I wanted to look at was if you looked at loans eligible for a low, balance pay up, if you looked at things that are also inherently

some call protection. are certain states that have, you know, refi taxes. credit stories such as lower FICO rent bands that people think have some type of, you know, call protection from a prepayment standpoint. There's also, you know, agency programs like Home Ready, Home Possible, Mission Score, that also get spec pay ups because of this, call protection that they're looking for, right? The idea that you need to have loans that won't prepay as fast as you think.

And so when we looked across the population of loans across our hedging clients, it was actually quite astonishing how much of the population of loans is eligible for some type of spec pay up. was at just a little under 70 % last month of loans that were eligible for a specified pay up. Now, the vast majority of that came from loans less than 350K, about 55 % of loans are actually less than 350K.

from a loan amount that we see in our hedging system. So all those loans are eligible for some type of pay up from an investor. And then you see things such as investments in second homes, right? of folks think that those loans have punitive LLPAs and we see some people pay up for them. You see FICO is less than 700. You see the mission score, home ready, home possible kind of make up that balance. But I think it's an interesting thing to track to see how much of the loans eligible in the secondary market to be sold.

actually are not going to be delivered into a TBA and potentially delivered into a spec pool instead. an interesting trend for us to watch, especially when it comes in regard to hedging instruments, because it could call on the question of sanctity of the TBA as that hedge that everyone uses from a risk management perspective.

Brennan O'Connell (: 10:23

Yeah, that's an interesting one. And when we probably will, you know, want to start tracking a little bit more closely the market advantage and on our reporting here that I've actually got two stats of the month here, if you'll indulge me. So one, more related to what we're talking about here on the refinance side, although we were up 70 % month over month here in August, we were actually down 10 % year over year. So, it seems like forever ago, but we were.

kind of like low sixes in terms of rates at, the fall of 24. And so, uh, we were actually down, year over year and, uh, September was a really strong month for refinances last year. And I think, we are about an eighth of a rate, uh, or an eighth of a point up we were in August of last year and about down 10%. So it gives us some amount

sort of ability to look into the crystal ball and say, well, you know, if we pick up another eighth, it, ⁓ you know, probably going to be another 10 % plus pickup, especially given the additional production that's occurred between, now and this time last year and the additional folks who are going to be in the money here. some more optimism if rate trends continue along this path. And then the other, bonus to the bonus stat.

which I can't help myself here, Vince Lombardi pressure behind me. Reggie White and one other player are the only two defensive ends in NFL history to have 12 or more sacks in their first four years playing in the NFL. The other individual is the recently acquired by the Greenback Packers, Michael Parsons. So I'll leave you both with that.

And we're very excited for the start of the NFL season here.

Olivia DeLancey (: 12:11

Brennan, did you know that stat from memory or did you look that up? I have to ask.

Brennan O'Connell (: 12:19

any packer stats like that. They're not in my memory, they're in my soul, Olivia.

Olivia DeLancey (: 12:25

Fair enough.

Mike Vough (: 12:27

As a Penn State guy, I'm rooting for you. Michael Parsons is a beast and Nittany Lions are number two in the country. So we could both be happy, Brandon.

Olivia DeLancey (: 12:35

All right, thanks guys.

Olivia DeLancey (: 12:37

Okay, it's time to welcome this month's guest. I'm so excited. We have our Chief Technology Officer, Siever Silliman, joining us. Siever, welcome to the Market Advantage.

Seever Sulaiman (: 12:46

Thank you. Thank you for having me.

Olivia DeLancey (: 12:48

Yeah, so we wanted to have you today because you just gave a really interesting presentation at HousingWire's AI Summit and your topic was how to measure and maximize ROI on your AI investments. So you're certainly one of our AI experts here at Optimal Blue and I thought your content was so good. I had the pleasure of being in the audience, seeing it real time. of wanted to dig a little deeper on some of the concepts you presented. So.

To kind of kick off, one thing that I thought was really interesting and that I could tell really resonated and got the audience thinking too was this concept of the AI hype cycle. Can you explain what that is for the audience?

Seever Sulaiman (: 13:30

Yeah, certainly. So that's actually a ⁓ that was drawn by Gartner. And it was specific to using Gen.ai for a system. I believe it applies to just using generative AI in general, not just for coding a system. And the way they described it is of a two dimensions, right? Time and expectation. So at the beginning of what they call the innovation trigger, where

us, Optimal Blue. In October: 2023

They had a peak of inflation of everything they can do. You can probably take a financial document and ⁓ decipher it and extract all the data elements out of it with no errors. So that's kind of the peak of inflation of inflated expectations as they called it. But slowly as we started experimenting with generative AI tools, with using LLMs directly or using some of the tools that were built on LLMs.

we realized, and I think the industry in general, not just mortgage industry, I think all users in general, all implementers, realized that it really isn't magic, right? It's really more, a very complex, intelligent, pattern matching algorithm that you can use in multiple use cases. So we came down to what Gartner calls the trough of disillusionment, right? Where we had a reality check, but then as we hit that sort of

bottom, rock bottom, we're starting to climb up to that slope of enlightenment because we now know what we can and cannot do with it. For example, we always talk about prompt engineering. When you prompt an LLM, the more information you provide it, whether it's examples, the formats you want, experience, the better the response you're going to get from it. So just by learning, by using and getting experience with using generative AI,

you're going to get a lot more out of it. So that's really that slope of enlightenment where you're to move up because you know what it cannot do, but you really know how to harness the power of generative AI, the LLMs, the tools that are built on top of LLMs to use cases out of it. And at the end, it kind of reaches that plateau of productivity, which I don't think any of us are there yet.

New versions of LLM come out, you know, GPT-5 was just out a few weeks ago. So as we learn, they learn. The model producers also learn from us, from the experiences that we're having with it, and they improve their models as well. So I thought it was very relevant to show and kind of show where people are in general in terms hype cycle.

Mike Vough (: 16:39

Steve, I thought that chart was super informative and I could see a lot of our kind of maturation through that process. of the things I wanted to like pick your brain on, and you mentioned during the conversation, was this idea of like a small model language, or a small language model instead of a large language model. Everybody keeps hearing about these large language models and more of how folks are figuring out how to optimize the operational side.

like what models fit in what spots and how folks are tweaking those dials. So maybe you can just go off on those two points for a minute or two here.

Seever Sulaiman (: 17:17

Yeah, absolutely. think

the context of that discussion was around ROI, The return on investment or return on intelligence. And how do you maximize your return? One way is to sort of minimize your investment, right? And the idea was you have to look at your use case for generative AI and not every use case requires a large language model, an LLM, right? You can actually potentially

satisfy your use case using a small language model. And the difference between an SLM and an LLM is simply the number of parameters that the model is trained on, the size. So literally small versus large. So for example, a small language model is trained on, could be 20 billion parameters. What that means is you have, think of it as you have a machine with 20 billion tiny small knobs and you're

adjusting those knobs to capacity of your machine and the intelligence of your machine. Compare that to one trillion for a GPT-4, which is a large language model. So there's no doubt that SLM is not to be as complex, is not going to be as accurate or as powerful as an SLM, but the small language model gives you lot of benefits. One is a lot faster.

Because the processing will be faster is trained on in a smaller the size of the static model much smaller ⁓ fewer parameters and You can also run on a machine so you can actually host it on your own cloud environment You can even host it in a laptop on a beefy laptop, and it doesn't require know the sort of large 200 GPUs that GPT-4 was trained on

The only difference really, really is just in the capacity, right, or the accuracy and the ability to process very large documents or very large information. For example, you could have a 20-page financial document that you want to feed to an LLM. You may not get the same result from an SLM. Now, I'll give you an example. last week about 5.3, for example. 5.3 is a Microsoft SLM. They went through 5.1, 5.2, and now 5.3.

And finally, there's a use case for agriculture. Farmers with no internet access, they actually have the ability, and they do this in India, they have the ability to use to build a complete model and a system around using that ⁓ SLM and they fine-tuned it based on the information that they fed it with agricultural information. And farmers are able to use it with no internet access on a small machine.

So it has its own use cases. satisfies that use case. It may not be enough for a complex product like Ask Obi, but it certainly is good enough for simpler use case that you just need language understanding or processing simple prompts or simple documents.

Mike Vough (: 20:21

Yes, so it basically comes down to the scale of the ask and the data digested, right? So if it's a very like down the middle like thing, like with not a lot of data, you could have some savings from a cost perspective on using an SLM versus an LLM. And any like feel for like, is there like a rule of thumb on like the cost savings between like an SLM or an LLM? Anything you could share on that front?

Seever Sulaiman (: 20:27

Exactly.

Yeah, I mean, the cost saving could be significant. So if we just talk about the cost of an LLM, for example, the cost of an LLM or large language model like GPT-4 or GPT-5, you cannot host on your laptop. I wouldn't recommend that you host it in your own cloud environment because you're going to need a lot of GPUs. So you naturally would use the one that is hosted by Microsoft.

Azure or other cloud environments. For those, you pay by token. And that token usage, unless you can predict it and you buy a reserved instance, it could get pretty pricey. With an SLM, if you're hosting it yourself, you don't pay anything. Because all you pay for is the machine that you're hosting it on. And some of them are open source that you can just bring in. I don't believe there is a license cost for a Pi 3. You download the model, you host it on your own machine.

your cost is virtually zero. While in LLM, you pay for tokens every time you're in a prom, you're paying for that.

Mike Vough (: 21:53

That's super interesting. it's almost like a true SAS model, where you can pay for it, download it, you can have access to it, use it as much as you want, as compared to model you mentioned for the large language models, where it's a fee plus usage, or per scale that you're paying for.

Seever Sulaiman (: 22:11

Exactly, you definitely pay for the usage of the LLM with tokens.

Mike Vough (: 22:16

There was that you brought up in your presentation that I just loved, especially like watching from the sidelines the optimal blue and being part of the beginning of getting some of this AI stuff off the ground here was the center of excellence. so was hoping that you could spend a little bit of the listeners through what the center of excellence concept means and then how we're how we're setting that up here at optimal blue.

Seever Sulaiman (: 22:40

Yeah, thank you for bringing that up. And yes, Mike, you were certainly one of the adopters of AI with an optimal blue. And we appreciate that. We talked about Compass Edge in our presentation at the end, and maybe we'll get to that. as far as set of excellence, so for me, it is very important that anything we do in technology, that we have a good base and good foundation. Just like when you build a new product, the first thing you want to do

is design and build architecture. And also when we embark on any new process, just like generative AI, you have to, in my opinion, create that center of excellence. And what that means is you bring in expertise and you also grow expertise within your own team. We don't just, so we did hire a couple of AI architects from the outside. We partnered with Microsoft that helped us with, as you know, trained us on co-pilot, trained us on...

machine learning on LLMs, et cetera. But we also created a architecture team that is AI specific within Optimal Blue. So we trained that team. We invested heavily, in myself, getting myself educated, my architects educated. Mike, you're part of some of those training that we've done. So create that center of excellence, meaning you create a staff that about

where AI is, where generative AI is, where it's going, how we implement, and then experimenting with different use cases. We're not going to get it right the first time because we can overuse it or underuse it. So that's what we've done. And then that also goes to the technical operation aspect. Our CIO and Core 5.16, also part of that center of excellence where they have AI experts within that environment, within that team.

so we can move forward and advance the usage and utilization of generative AI as a whole in the company. And that's what I would recommend for any company that was, I mean, the center of excellence could be one or two people, but you got to have the time and the investment in those people to learn and give them the time to stay up to speed, go to training, go to the Microsoft events or the Amazon events, be part of webinars, read on what's coming next and...

allow time to experiment with some of these new tools and new LLMs to see what we can do.

Mike Vough (: 25:04

Yeah, I was just so happy to see that because I think it has been 18 months in the making of setting up that center of excellence here. And you mentioned the great partnership we have with Microsoft. that helped us get kick started here. But we're also setting up our own foundation and theories here on how to best deploy this stuff. And I think it culminates with hackathon that we had a couple of months back. There were just some really amazing use cases that were starting to come to life there.

just super bullish on all the things that we're doing here about AI. Another question, Zevers, to keep drilling home on your presentation was this idea of AI personas and the categories around there. Maybe you could just take us through what some of those categories are and where you see people fit. I saw that optimal blue fit into a couple of different spots there the presentation. So maybe you could just take us through that.

Seever Sulaiman (: 25:49

Yeah.

Yeah, certainly. And it sounds like, you certainly were listening to the presentation. So thank you. So the way I thought about presenting ROI, you kind of first have to talk about what is your role as a company in those. And I simplified into kind of three layers. At the bottom, you have the model producers, the people who actually build the models. OpenAI, Google Gemini, Claude with Sonnet, Anthropic Claude.

Microsoft, of course, we mentioned with SLMs. that's kind of the model producer. And just on that, think it's flattering when someone calls me or our colleagues that were at the housing wire, AI experts. We really aren't, right? We're not AI experts. We're not the ones who built the generative AI models. We're becoming experts at using the generative AI models, right? That's really where our expertise come in.

I we understand how it was built and what it is good for, expertise really is more in the utilization of these tools so they can benefit our products, benefit our teams, and ultimately benefit our clients. But anyway, so back to the stack of the three-layer stack, at the bottom you have model producers, right? And then on the middle tier, you have the implementers. The implementer are those who actually kind of, think about a low-level implementation of

the LLM directly in an application. And that's what Optimal Blue does. We use the LLMs in our Ask Obi, in our originator assistant. We use them in Compass Edge Assistance. So we go directly to the LLM, post it in the cloud, right? Because we're not buying 200 GPUs. And that's the implementers. And we have some lenders who also are implementers, right? And you have other technology providers like us that could be also implementers.

Clearly, OpenAI themselves implement an LLM in the chat GPT application. Chat GPT is an application that uses a GPT model. Same with Microsoft Copilot. They also use an LLM. And then at the very top of the three-stack layer is the users, the end user. So a mortgage lender who uses Optimal Blue's applications is an indirect user of generative AI LLMs because we implemented it. They're using our products, so they become an end user.

the blue ourselves, we use Microsoft Office 365 Copilot. So we're an end user of the implementer's application that uses the LLM. Of course, the world, everybody uses chat GPT or some other form of chat application based on an LLM. So everybody's an end user if you think about

Olivia DeLancey (: 28:40

So Sivar, the crux of your presentation was making sure you're maximizing ROI with the AI you're using. So if I'm an end user, and to your point, we're all end users of AI, but particularly if I'm a mortgage lender using an AI tool, and I don't have any idea where to start in terms of measuring my ROI, what would your recommendation be?

Seever Sulaiman (: 29:05

Certainly, and just on the end user, just sort of a side note, unrelated to the business, if you, as all of us, as end users, as people sitting at home using ChatGPT or Gemini, it's now part of Google, how do we maximize our own ROI on using ChatGPT? It's my time spent typing a search.

The way I could maximize it is by actually giving it a good prompt, right? Not just use it as a Google search. And you see this, you know, posts on LinkedIn and everywhere else, which is very true that every, most of us, including myself, we're using chat GPT as a Google search, but that's really not what it's built for. And we're actually costing OpenAI a lot of money, by the way. We didn't, we don't know that, but every word we type in there, it's costing them money. But the way we would maximize as an end user, our

own utilization of it is really by using it and giving it prompts that really matter for us so we can get the answers. But back to the mortgage lenders and mortgage clients, I think the way you maximize it is first you have to understand what is it that you're looking for? What is your use case and what is the best? I talked about one of the items as far as your investment is in identifying high impact use cases. And that's what we did. And actually with Mike,

Vogue himself, about 18 months ago, we were new to generative AI. We started about two years ago. And we said, OK, what do we use it for? So we spent the time. We invested optimal lose time in identifying high-impact use cases for generative AI. And the result, the outcome was Compaset's assistance, which is, if you saw the presentation at the end, we had a case study with our of Omaha.

that can't live without that information anymore. So that's really how you would maximize it by identifying number one, high impact use cases. And then the other one is to understand your return, you have to understand your cost. So you have to baseline where you are. That baseline could be in productivity. For example, if I have my development team using GitHub Copilot for productivity, I'm baselining my development velocity today.

and I'm going to measure it in six months or three months or whatever to see what the difference is. Similarly, if you're a mortgage lender and you're doing, let's just talk about hedging, right? You baseline the time that you spend per loan, it's dollar amount or number of loans, before using, for example, Optimal Blue's Compass Edge Assistance, and after you use them, and you look at how much time it's saving you, that's your return. That's how you maximize your return.

A, identify your high-intact use cases, and of course, by utilizing tools that are proven and tested. And you would let providers or vendors like us, whose job is to invest in AI, for you, the lender, and trust us to deliver those type of tools for you that maximizes your return on time, whether that time is invested in AI or using our product.

Mike Vough (: 32:19

Yeah, just to build off that, especially with the Compass Edge assistance, we spent time looking at what our desk and then what our clients did monotonously every day. like, you know, the write ups that come out of Compass Edge today, that detail, the top five reasons why your position changed or why you made or lost money. That was something, repetitive task that somebody had to do every day. And sometimes it took five minutes, but there were those times where it take two hours to figure out the reason why. And so.

And then imagine now doing that at scale across the product.

It used to be exemplary customer service or what the master user would be able to do. And now it's just baseline product functionality, right? It's an automatic email that comes out every morning that details the whys and hows. And there was a ton of iteration and I got pretty good at writing these really long prompts that like I sometimes I had to get, had to like bring them down to manage tokens, et cetera. it provided that like that change from like what I would call like exemplary service.

to like a baseline product feature changes that give people back time in their day. So they could spend it on a strategy, right? You can spend it on figuring out like, how could I hedge more effectively or what, what a different investor mix do I need to use? Do I need to stop using this different cross-hedge strategy? That's not working. It allowed people more time to spend on these very strategic, kind of like deep thinking projects as opposed to like, let me like knock out, you know, 50 items on the to-do list.

and do that because it's kind of easier as opposed to spending time on like the harder deep thinking problem.

Seever Sulaiman (: 33:59

You me.

Olivia DeLancey (: 34:00

While we're talking about client benefits and client use cases, actually a question for either one of you, are there any other interesting use cases you've seen of ways our clients have been realizing ROI with our AI products?

Seever Sulaiman (: 34:14

Yeah, I think, for example, we just launched, just six months ago in our summit, our originator assistant, right? And the originator assistant, which is a sort of complex workflow of identifying additional products that a borrower can be eligible for. So the loan officer is looking at our PPA application and they have a loan scenario. The originator assistant is giving them the option of

offering additional products and better pricing to the borrower if they improve upon certain parameters and minor small improvements on FICO and putting a little more down payment, $5,000 more in down payment, et cetera. So that has actually given the loan operators the ability to serve their borrowers in a better way because they get a loan scenario, they get the products, they just do a normal PPE search.

product of pricing engine search, but the originator assistant is giving them more options. So when a loan officer is able to satisfy their customers, which is the borrower, and give them more, they're going to repeat customers. Hopefully, they're going to get word of mouth and get more and more borrowers and be able to generate more loans.

Mike Vough (: 35:29

Yeah, I love that feature. The one I'm going to call it because we have so many right now is Ask Obi. Like I just think that that brings back so much time into people's days, especially at the executive class of people who are typically like in a meeting and they just need that answer right now.

or they need the chart formatted for their PowerPoint that they're working on and it's eight o'clock at night and no one's around, it kind of opens up that like anytime, anywhere access to data without having to know SQL queries. I definitely love a SQL query, but like it's not the same for everybody. Not everybody knows that language or understands that, even though a chat GPT makes it really easy. But ASCII OB is that perfect like anytime assistant that allows folks to just make those decisions just way faster.

Seever Sulaiman (: 36:13

Exactly. And Ask Ovi would say, our probably most complex AI product. And it's not one that we can use in SLM for sure. It uses nine LLM where it's an agentic workflow that uses an LLM at multiple points. And because it's so complex, we basically use our entire data set to be able to answer those questions and identifying intent of the question.

So it's very complex and that's a really good use case for when you would want to invest in a large language model.

Olivia DeLancey (: 36:46

Perfect. Siever, this feels like a good moment to pose our bonus question to you. So we ask every guest this question to close us out. What is one thing in your opinion that you think lenders should be doing in today's market to maximize their profitability?

Seever Sulaiman (: 37:02

I'm going to ask you that question specific to generative AI because there are many ways you can do that if you're, especially if you're a lender using Optimal Blue and of course I'm biased, you being with Optimal Blue, but specific to generative AI, I think you have to know where you are in that hype cycle, right? And if you are at the very beginning, you...

you have to know that you're going to invest some time and money in training and building that center of excellence. And as I mentioned in the presentation, the good news for those that have not started with generative AI is they can basically skip, go from number one to number four, meaning skip the peak of inflated expectations and skip the disillusionment and the disappointment and go to the slope of enlightenment. So you can skip those by learning from the rest of us.

And of course, all the blogs and posts that you see on LinkedIn and other media. But that's what I would think that if you are beginning with that, you can learn from others. But if you're doing a proof of concept, you can certainly make it a very low resource one with one or two people and really stay up to speed with what's coming up. There are a lot of tools. there's so many that you have to filter out what is good and what's bad.

that's where the center of excellence becomes important, because you have to have a team that is dedicated. It could be one person that is dedicated to identifying what is a good tool or what's a good LLM for my use case and for my company. So I answer that specifically with generative AI. That's how I think you can maximize your profitability on using it. And of course, like I said earlier, I mean, we

Our business as technology providers is to invest in technology, is to invest in generative AI. Our center of excellence is going to be much bigger than the center of excellence of a lender whose business is willing to generate and make loans, but rely on Optimal Blue to be that technology provider and stay up ahead of innovation and generative AI. So you can always look to Optimal Blue for using, you know...

Genetic AI and advanced technology to maximize your profit.

Olivia DeLancey (: 39:21

Great answer. Thank you, Siever. And I'll be sure to include a link to the hype cycle chart in our show notes so that people can visualize what you're referring to and maybe decide where they land, where their organization lands. But Siever, it's been such a pleasure to have you. Thank you so much for joining us and hope you'll come back again soon.

Seever Sulaiman (: 39:42

Thank very much. This was terrific. I enjoyed it.