National Mortgage News' Maria Volkova speaks with Ziggy Jonsson, SVP of engineering at Better, about innovative technologies at his company and his future outlook on technology.
Transcription:
Transcripts are generated using a combination of speech recognition software and human transcribers, and may contain errors. Please check the corresponding audio for the authoritative record.
Maria Volkova (00:10):
Hi everyone, my name is Maria Volkova and I'm a reporter at National Mortgage News. Today joining me for a brief discussion about artificial intelligence is Ziggy Jonsson. He's the Senior Vice President of Engineering at Better. Ziggy welcome.
Ziggy Jonsson (00:31):
Thank you. And thank you for having me, Maria.
Maria Volkova (00:34):
So to kick off our conversation about artificial intelligence and innovation at Better, I was hoping you could give us a brief introduction of yourself to our audience. What does it mean to be an SVP of engineering at better?
Ziggy Jonsson (00:54):
Yeah, I can do that. So I run engineering at better. So when you think about technology split into product and engineering, and my job is probably a little bit different than many people leading engineering because we have high velocity, we need involvement in all places and we are moving things along pretty fast. So in addition to just managing the teams, I'm also heavily involved in feature development across the company.
Maria Volkova (01:23):
Okay. So thanks for that. So the last time that we spoke was in September at National Mortgage News's Digo, and you were on my panel and we were discussing generative AI and how it may or may not disrupt the mortgage industry. And I wanted to kick off our conversation and ask you, so it's a new year. Do you think hype around generative AI is still valid today?
Ziggy Jonsson (02:01):
I think it's even more valid now than it was when you last spoke. I know it's been only a few months, but the last year development in the space is unbelievable. And I think it is going to change a lot of things in terms of communication with companies and I think in markets in particular because markets has this vast set of regulations, vast set of criteria and all this stuff. And then on the other end you have the customer who's just trying to navigate, how do I get my house? I think there are unbelievable opportunities in using gen AI to making sure that the customer is informed, understands and can make the right choices when they pursue their home buying journey.
Maria Volkova (02:45):
In your opinion, very briefly, what operational hurdles in the mortgage process could generative AI fix and maybe how can it simplify the roles of loan officers, underwriters, processors? It seems like there's an overall sentiment in the industry that investment into AI is worthwhile. So it would be wonderful to hear your thoughts about what it could potentially simplify.
Ziggy Jonsson (03:19):
So it's a good question, but when you look at ai, AI is not new. It's been used for a long time. We've been investing in AI for a long time. I think that what's really new here is the generative ai, and that's where it translates from neural nets to English and back from English back again, which has been the missing piece to be able to have conversations with a customer or help an LO understand a very complicated situation in a quick summary that kind of explains what the issues are. But when you think about ai, AI at the core is the machine making decisions for people or following flows. So that's something that we've been investing in. And then Betsy, our AI product comes from top as a layer that connects humans to all the actual artificial intelligence that's happening under the covers.
Maria Volkova (04:11):
Right. So you bring up Betsy. Yeah, that was kind of my lead in terms of how is generative AI fixing some of the problems in the mortgage space. So better a few months ago launched officially Betsy and Betsy is an AI voice bot. And to talk about Betsy, first off, I wanted to ask why did better opt for a voice assistant versus an AI chat bot? I know that a lot of kind of mortgage lender competitors have launched very actively chatbots, but you guys went the AI voice assistant route. So would be, I'm curious to hear your
Ziggy Jonsson (05:00):
Thoughts. That's a very good question. So our experience with chatbots is not great, and we also know particularly in purchase, people want to be on the phone. It is not only calls, it's texting as well, but I think in some cases people don't trust the chatbot or think it's just has rudimentary answers. But they have a house they want to close and they want to talk to somebody knowledgeable. So we actually, a significant part of the mortgage journey for a mortgage company is people talking to customers, right? And customers have questions they need to be able to pick between products, there's something they need to change on the file. So there's a big human element in the entire mortgage journey, and we actually feel strongly that whichever channel customers want to communicate through, we should be able to address any concerns and make any changes they need along the way and stick to them 24 7. And one of the best ways to do that is through telephony. And keep in mind telephony is just a communication method to the same brain. So Betsy can talk on any channel, but we started with telephony because that was the most meaningful way to get to consumers and be able to help them where they are in their journey.
Maria Volkova (06:24):
So you mentioned that, have you guys tested AI chatbots prior to kind of going the AI voice spot route?
Ziggy Jonsson (06:36):
So we have tried quite many things throughout time. For us it was a unique opportunity to tackle the main channel that people communicate and push that through. And Betsy has taken over oun of our calls very successfully. And in many cases Betsy ends up transferring you to a licensed loan officer if it's a licensed activity you need to talk about. But if it's questions that Betsy can answer, then she can just resolve the customer's questions or worries on the phone without even having to loop in a human.
Maria Volkova (07:14):
So I was hoping you could expand a little bit upon what tangible issues is the voice assistant solving and also what generative AI component is in what makes Betsy generative ai, does it change its answers? If you could maybe comment
Ziggy Jonsson (07:39):
On that. Yeah, obviously this could go very technical, but let me see if I can give a, so most solutions that are generative ai, they're based on one of the foundational models like either open AI or Anthropic or any of the other guys because creating a foundational model is expensive, but then where the solutioning comes is on top of a foundational model. So it has to do with fine tuning models, which is fine tuning layers on top of those models. It has to do with having knowledge basis that actually return the correct context that the LLM needs to evaluate. And then finally it has to do with prompting and function calling. So actually advancing a product in ai, you really must get all of those components correct to be able to kind of be confident that the AI is actually responding with the right answer.
(08:39):
And when the AI doesn't know the answer, it doesn't make up an answer, but actually get you to human whenever required. We are pushing the envelope on getting Betsy to do more and more things and I think that's where the investment path in AI is going to be for most companies is what are the roles for a human just needs a resolution to something and the best way to communicate it is English and it can be resolved in half a minute instead of being on hold for 10 minutes and having a person call back. So that's kind of the direction we are going in.
Maria Volkova (09:15):
And do you guys have any future plans for Betsy? Apart from it, just answering very basic customer inquiries and transferring a call if it can't give an answer.
Ziggy Jonsson (09:32):
So obviously I think there's some detail in our release around Betsy. I think Vishal mentioned publicly that in five or 10 years, 99% of all tasks that need to get done would be done by bots. And what we want to do is there are some things that humans are going to do, but everything that a human doesn't have to do and can be done by the machine, it benefits everybody. And if you're a customer that don't want to deal with machines, deal with humans, that's fine. But if your core objective is to get the ref refunders done or get the maximum amount from your home equity or get the next house and you just care about getting where you want to go, then we want to make sure we have all the tools at the disposal to add anything you need. And that's what we've been doing outside of gen ai. Gen AI is kind of a sliver on top that helps us to do even better,
Maria Volkova (10:28):
Better. So Betsy is powered by your proprietary system, your LOS Tinman, and I was hoping you could tell our audience a little bit more about tinman. What else does it power within the system? What are some of the tasks that it helps loan officers do@better.com?
Ziggy Jonsson (10:54):
Yeah, very interesting question. So I think this is one of the key differentiators for better is that when we started the company, we actually started writing our own loan origination system. The way we've developed the system is probably a little bit different than the big LOS you have in play because again, we wanted to have AI be applied so that people don't need to know what they have to do next or think about what is outstanding. We wanted the machine to deliver tasks whether a customer has to do something, a loan officer has to do something or closer has to do something. And as you know, markets is a very complicated, so each individual markets application flow varies because there are different permutations of the paths that a borrower needs to take from when pre-approval started until the loan is funded. So the tinman LOS, it's basically a big graph of for any permutation of final loan product, what are the things under what circumstances do, what things need to be done and by whom.
(12:01):
And it's basically the machine that's driving the entire flow. Also, one big differentiator, 10 minutes is a single system. So most mortgage companies, they have fragmented system. They may have one system that's tackling pre-approvals, they may have one LOS, they may have one CRM, they may have one software for appraisal and it just needs human involvement throughout the entire process. And more importantly, it's very hard to have full transparency to the customer. Usually you have to call and talk and you have to log into system, figure it out. So what differentiates Tim, and it's a single system. So for a given loan file, we have a snapshot at any given moment in time of every single factual piece of data around that loan file, whether it's self-reported numbers, verified numbers like calculations of ratios, all the stuff. And as things change, the system is changing with it.
(13:02):
So did we have a change of circumstances that needs a new loan that's sent out, et cetera, et cetera. It's basically the machine driving all of this. And it's interesting when you think about gen ai, it's a unique strength to better that we have our own LOS and everything is in a single place because when we are feeding gen AI with factual data, here are the things outstanding on the file, here's all the factual data, having it all in a single place in A JSO BLO that can ingest into the LLM, it's the best way to get to give the LLM full context of everything that's going on. So we've been very unfortunate to be in a position that we can drive Betsy behavior by our own LOS because we control both parts of it. It's very easy for us to push forward and kind of improve features.
Maria Volkova (14:00):
So just to make sure that I understand, you guys use AI to I guess analyze data inside of tinman and that helps Betsy give answers?
Ziggy Jonsson (14:13):
Yeah, so that's one, but it's interesting. That's why when I talk about gen ai, I talk about it as a kind of sliver on top because it's interesting if a customer needs something or an internal person needs something, they need to figure something out. The interface that they use is like tinman, it is the l os, right? You use a mouse and a keyboard and you need to know you can't have a conversation with a platform. You have to go to the things you want to do and click. So where Betsy comes on top is you want to achieve something or your question about something, see it has the same access to tinman as a user, but C actually figure, okay, I need to look at this and then give you an answer based on that. Plus a knowledge base that accompanies everything. It's like, okay, with this factual data and those outstanding activities and this information from the knowledge base, how do you combine that into an answer to the customer or to an internal person?
Maria Volkova (15:13):
I'm also curious, we discussed this previously, but you guys built very much an integrated system and it does have the potential of not only being limited to mortgage lending for example. I was wondering if you'd like to provide any commentary about are there plans to potentially expand into other financial products?
Ziggy Jonsson (15:41):
So Tinman is being applied to mortgages. Tinman is built in a way where it is a loan origination system and because we built it, there's nothing that confines it to markets except the fact that the graphs of activities that we built, they're all related to markets. So a few years back we started doing HELOCs. So we actually very quickly from making the decision. Now HELOC is also underwritten by Tinman, even though it's very related to the conforming markets, it is a different path. I don't think this is the venue to talk about company plans that haven't been discussed, but from a systems perspective, tinman remains a huge asset to the company I think because it gives us the flexibility to make any decisions in the future.
Maria Volkova (16:35):
There's a lot of potential there. So when you guys are developing tech products or thinking about implementing tech products, do you guys build everything in-house? If not, how do you decide whether to build things in-house versus use vendors to help your operations along?
Ziggy Jonsson (17:05):
That's a great question. That built versus buy, and we are faced with that every day. I mean obviously every single case is different, but in the end it's like ROI, which option? So in my personal view that when approach for this is like do we need something that's already solved by the industry and everybody's using standard solutions. In that case we're not building an enterprise value but trying to replicate that in-house. But many cases where when we are building something, it's something new, it's something proprietary, and in those cases it usually doesn't make sense to necessarily call the vendors. One example is we have our own pricing engine, which is very rare for mortgage companies. It's actually part of tinman and that's a thing that we could use vendors, but then we have less control over the speed of our pricing of how we can quickly update the pricing sheets, et cetera, et cetera. This is an example where we built this in-house and that engine gives you pricing to complex pricing questions in few hundred milliseconds and some things that are very core to our business is very good to control because then we control the quality and the latency as well, which is very important.
Maria Volkova (18:29):
And I have another question that I wanted to ask you. When you guys are developing Gen AI products in-house, what are you being mindful of? I know there's a lot of concern around gen ai, maybe biases, faulty data that's being used. What are you thinking about when you're building out these products?
Ziggy Jonsson (18:57):
Yeah, it's interesting. I'm thinking back to the FSFA hackathon that we talked about when you spoke last time. It is a regulated industry that we're in. So it is imperative that whatever gen AI is speaking to customers about is accurate and is correct. And it kind of keeps to things Gen A can talk about and doesn't talk about all the things from our perspective and just generically, how do you do that? So with I, there's always an element of risk, right? Because you can't control every conversation, but there are mitigating factors that you can apply. For example, in Gen I, particularly in telephony, latency matters a lot. So you can't have a model that's thinking through for 10 seconds then responding. But what you can do is you can have a heavier model on the site that's monitoring the conversation as it happens and can actually make any corrections if anything is going in the wrong direction or anything has to be explained better.
(20:00):
You can have that model do that. And there's another component that I think is important and it's actually test suites. It's a very common thing in AI where you're solving for this one thing, but then another thing that you had tackled last week is now not working as it should. So actually having test suites for all changes to making sure that any cases where AI could go off the rail, we test that it doesn't. That's actually very important to be able to push development forward. But yeah, it's a super exciting space, but production is a little bit different than POCs or being at your own computer. But so far the experience has been really good.
Maria Volkova (20:47):
So you mentioned regulation and the mortgage industry is very much heavily regulated, but what makes this year unique is that a new administration will take office come January 20th I believe. And with that some regulations may ease if that is the case. How might a more relaxed regulatory environment impact the development of technology in the mortgage industry? I know there's a lot of concerns by stakeholders in terms of using Gen E AI tools because they're uncertain about what the regulations are, but times are changing maybe. Yeah. What are your thoughts about that?
Ziggy Jonsson (21:41):
So from the start, from the conception of better day, there's been a very key focus on the customer on saving the customer money, on giving customer the best experience. So when you look at the regulation in housing in general, obviously there are some regulations that are there to protect the customer from something bad we experienced and we don't want to happen again. But then there may be some regulation that's actually more geared towards protecting interest of lobbying organizations that are kind of taking a piece of the cut. I think if you're going to see things simplified for the customer but still keep all the customer protection in place, I think it'll be a big win. I think any regulation changes that are geared towards the consumer are going to be really good for us because what we try to do is our technology is kind of dynamic, so we actually adjust how Tinman operates with changing circumstances. So hopefully we're going to see some positive changes. We look forward to it, most certainly.
Maria Volkova (22:51):
And you mentioned that FHFA sprint, so my last question is that there are some murmurs that maybe Fannie Mae and Freddie Mac may be released from conservatorship, possibly, who knows? We'll see. But if that's the case, and it seems like Fannie and Freddie have cautiously, but they've led a lot of mortgage tech innovation in our space. Do you expect their hypothetical release to have an impact on how Mortgage Tech develops?
Ziggy Jonsson (23:27):
I hope so. I hope so. I think so. Fannie and Freddie have been forward thinking in the current state, so I'm hoping if they go more private, the growth opportunities are on tech and are on advancing their underwriting systems. I think when you think about a typical mortgage criteria, it's very fixed. You have all those sheets and it's a long nested list of things that had to be exactly as there are. At the core of everything is fico. FICO is just a measure of credit worthiness, so this is something that's been very hard to solve for. There are plenty of people who are great borrowers, but their FICO scores aren't matching that. New borrowers that haven't had a lot of experience, but they have good jobs and nice salaries, et cetera, et cetera. Also, there are cases where people seem to have high credit card debt, but they're just paying off their cards every month, right? It's very different than somebody sitting with this. So where I see things going hopefully is that the credit policies are still protecting the customer, but also not denying people that realistically would be great borrowers. So I'm hoping we're going to see the market move in that direction more.
Maria Volkova (24:47):
Okay. Awesome. And the last question, we only have about two minutes left. Are there any new or cool technologies on the horizon that maybe you, you've seen or heard about that maybe the mortgage industry isn't using yet, but there's potential there?
Ziggy Jonsson (25:10):
It's a very good question. So outside of Gen ai, I think mortgage is complicated and I think the way mortgage companies are constructed and the way use technology is very fragmented, it's very expensive and none of it is geared to the customer. Most of us are geared to serving the needs of the people that work there, but they also, they only know what they know. They don't know how things could be different. I think we are going see, I mean better has certainly been at the forefront of this, but we are going to see more technology investments that are really helping the customer. The Gen AI is one part of it, but also just if you're a customer, why are you asked to do all those things? We strive to minimize the things that the customer has to do, but it's like getting to a place where getting a mortgage, we should realistically be able to piece everything together for you if you want to buy this house.
(26:07):
If it can get to a place where technology is resolving everything and it's like you're good to go and speaking from better, one of the things that we delivered I think last year is the one thing markets from the time you apply until, if you go through the flow, well, from the time you lock a loan, we give you a conditional approval within 24 hours. So within 24 hours you actually are fully confident in a transaction if we can squeeze everything together. So it's more like the markets become more like an afterthought. I think that will be the north star for the industry. And I just think the whole system in the US is way too expensive and technology should be solving that. That's our mission.
Maria Volkova (26:51):
Okay. Well on that note, Ziggy, thank you so much for joining this conversation and thank you to our audience for tuning in.
