Peter Wyman, CIO at Brundage Management Discusses Mastering Analytics and The Power of Decisioning Analytics Pro

SPEAKERS

Rich Alterman, Peter Wyman, Pia Wilson, Carl Spilker

Pia Wilson  00:20

Good afternoon everybody looks like I can see several people logging into the webinar so we're just going to give it a couple more minutes. While we wait for everyone to dial in and join into the webinar, you are automatically muted I did just want to let everyone know that so no need to worry about that we will have a live q&a at the end where we can you can either enter your question in chat or I can unmute you just know that that is coming in available towards the end of the presentation. Okay, so it looks like we are right at one o'clock. I think we'll just go ahead and, and get going. I know several other people might be joining. But I just want to be cognizant of time, obviously since we've only got an hour for this meeting, and we'll have the recording available to all afterwards. So that being said, I did want to just say good afternoon ladies and gentlemen. Welcome to our first GDS Lync client connect, mastering the art of analytics. I'm your host, Peter Wilson, and I'm absolutely delighted to have you all join us today. I wanted to start with introducing our distinguished guests. Joining us today is Peter Wyman, the Chief Information Officer of Brundage management company. We're also honored to have Carl Spilker GDS Lynx, Executive Vice President of analytics, and our moderator for this session is rich Ultraman, Executive Vice President of partnerships at GDS link. But before we plunge into the heart of our discussion, let's take a moment to outline what's in store for today. Brundage management has confronted many challenges from monitoring existing models to assessing borrower risks and streamlining their loan approval process. We'll delve into these hurdles and how GDS links cutting edge analytics solutions played a pivotal role in transforming transforming their business. Now, without further ado, let me hand over the reins to rich Alterman, who will skillfully guide our conversation and an earth the secrets to mastering analytics in today's ever evolving landscape. Rich over to you.

 

Rich Alterman  04:46

Thank you Peter, for the introduction and thank you everybody today all of our guests for joining GDSL Lynx first client connect and we hope you enjoy today's session and will join us for other sessions in the future. Thank you of course Peter for being our guest star today. And Carl as well for joining us this afternoon. So I'd like to start before we we touched on Peter, Carl, if you don't mind, I'd like to have you spend just a few minutes given a brief overview of your role at GDS link and the services that you deliver to our clients.

 

Carl Spilker  05:16

Yeah, absolutely. Thanks, Rich. Yeah, my role inside of Gds is heading up the analytics and advisory piece of the business, leveraging my 25 plus years as a chief risk chief credit officer out in the industry, and then building the tools inside of Gds to help us enhance and provide value for our clients. So been here with GDS coming close to four years.

 

Rich Alterman  05:42

Thanks for sharing that, Carl. Okay, Peter. So let's get started. Why don't we start off Peter with you given a little bit of background on Brundage management. And also, in particular, the Sanlam. Any business that you help operate

 

Peter Wyman  05:57

would be very happy to by way of introduction. I'm the Chief Information Officer at Brundage, and I've been there for almost almost seven years now. Brundage has been around a good deal longer. And the salon business has been in business for over 30 years. Brundage management is a privately family owned business, and it is the sole owner of some loan. So loan is in the business of making non secured consumer installment loans. And we do business through 214 branches. In seven different states, we're in to 36% states, those would be Illinois and New Mexico. And then we're in five non 36%. States. 36% is the ceiling for interest rates in those two states. The other five are not 36% of those would be Texas, Oklahoma, Missouri, Alabama, and Nevada. So we're spread around a bit. Shall I just

 

Rich Alterman  07:04

don't I mean, yeah. Yeah. So why don't we actually start by taking a little time travel? And let's kind of go back to 2018. And maybe you can start off by sharing, you know, what were some of the standard standard operating procedures at the time for your lending business. And some of the challenges, I understand that you had a lot of disparate data, and need to find a way to bring that together. So why don't we kind of start there, Peter, if it's okay,

 

Peter Wyman  07:30

we had a lot of challenges, I'll just say that the the business had been doing good business very successfully, for as I said, over 30 years, but hadn't hadn't adapted very well to changing times. They had some web assets, but basically, it was brochure where, and sometimes marketing and also consumer education materials. Each of the branches had their own server and their branches. So I had over 200 At that point, but 250 data islands with all this data spread everywhere. And we didn't have an online digital presence for purposes of originating loans. So everything was either on the phone or through the front door in each of the branches. And the processes were, for the most part manual with the exception, obviously, of the origination and servicing system itself.

 

Rich Alterman  08:25

So some of the challenges in a manual operation. You know, I think companies would often say if a consumer walked into any one branch, would they necessarily get the same decision? And, you know, what were some of those challenges that you might have faced? Well, I'll start

 

Peter Wyman  08:43

with the most obvious, which is, underwriting loans manually is a slow time consuming and somewhat invasive process on the part of the as it affects the customer. We at that point, we're actually calculating monthly operating budget for the consumer to try and get an idea of whether they had enough free income to repay the loan. Affordability is kind of an important aspect of the whole to the whole equation as far as we were concerned. So that process is pretty invasive. We're going to ask you about your grocery bill every month and things like that, that you probably don't know off the top of your head and probably don't feel really great about sharing all that information. So it's really not very client friendly. And to your earlier point, it is potentially subject to human error. Some people will react different ways to different people, different different manners and behaviors. And so, while we were very, very deliberate about our standardized training, and all of our operating procedures being standard and consistent, it's hard when you throw in the human factor, it's hard to say that everyone would have given the same answer to the same question had they walked into their branch But perhaps the most significant thing is it's not very scalable, right? It's certainly not suitable for the digital channel. Right.

 

Rich Alterman  10:09

So I know that we were, we're fortunate enough to win your business back in late 2019. When you decide to move forward with GDS Modelica, maybe you can talk a little bit about the journey to get to that point, you know, putting out an RFP, and you know, what were some of the key things that you were looking for in a platform like ours?

 

Peter Wyman  10:29

Well, we were very, very deliberate in our approach to looking for a vendor, we didn't have any preconceived ideas about who we would select, to begin with, looked at a number of organizations and did not send them the RFP, but wound up sending it to five different organizations, yourselves included. The RFP was very extensive, including, I can't remember exactly, but about 30 pages worth of here's how we think it should be built. That was entirely science fiction. It was, you know, I wrote up requirements as if I knew what I was talking about. But but the point was to engage each of the vendors in Alright, here's what we thought we wanted. How would you do this? And what would you suggest, and we got great advice. And counsel. The reason that we selected GDS, and I'm not sure I ever told you guys this is you had a combination of executional capability, coupled with the ability to help us build the necessary automation and the underwriting processes, we had already come to the conclusion that there was no way we could simply automate what we were doing currently, what we were doing currently just wouldn't work online. And as I said, was was very invasive and time consuming. So it was a fresh start. And we needed somebody with fresh ideas. That's how we selected you.

 

Rich Alterman  11:51

Okay, good. Well, let's, let's get back and do a little time travel moving forward now. So you get to a point where you decide that, hey, it might make sense to look at using analytics to build some consistent models. So maybe you kind of talk about, you know, what was some of the internal discussions going on? What were some of the drivers behind, getting to the point where you felt, you know, starting to look for companies and obviously, employing Carl and his services? How did that all kind of come about?

 

Peter Wyman  12:19

We knew that we did not have the internal skills to develop our own approach to defining the models. We felt like we needed a model for credit risk and another model for fraud risk. But we were not we knew we were not the right people to write those models. We were going to have to have somebody do it. And Carl and his team came across as very knowledgeable with a clear idea about how to go about doing that. And in fact, they jumped right in, we actually started with the model building, finished all the models and then said, Okay, now let's go build the the delivery mechanisms. And to Karl's credit, and his team's credit, they pretty much boil the ocean. They look at many different data providers, and helped us understand what made one set of data better in terms of its ability to predict risk than other sets of data. They then went through and did another filter that said, Okay, none of this stuff is free. How do we get the most bang for the buck? We wound up with two default models, one for new loan customers, and a second for former borrowers because they have very different credit risks.

 

Rich Alterman  13:41

Let's take a beat, if you don't mind, let's take a pause there. Okay. And maybe, Carl, maybe you can kind of just jump in here a little bit and give a little more insight on on, you know, the process, you all went through how you leverage machine learning technology, you know, Peter talked about all the different types of data sources you looked at. We don't need to, you know, talk about specific bureaus in general, but you know, how many did you look at? And, you know, what was the kind of timeframes to pull off together? And what are the what are the Peter mentioned, couple models, you know, what are they actually predicting?

 

Carl Spilker  14:14

Yeah, what was exciting about working with Peter and his team was how innovative they were on investigating the data sources, alternative data sources, the multiple bureaus and multiple bureaus services, they provided various scores. I want to say we evaluated up to eight to nine different data vendors. So you're talking 1000s upon 1000s of features, including an own internal set of Gds attributes. And what was really good about this was it was very transparent. We're very open as we build. I want to say we ended up building 22 to 25 different machine learning models, all the data, all the data minus, just each by themselves. calls. But the idea was to drive that maximum lift in predictive value for default. And we ended up embedding down what would be called a never pay plus a true charge off variable into the new lending and the former borrower. And what was nice as the Brundidge, existing clients had a lot of loyalty, sticking with the clients and perform really well. And new loans, like everybody in the industry know, you're kind of making that initial bet to you build that relationship. So we had a separate model signaling that. So the idea was to target those never pay plus aging defaults into into the model. So you captured both of those. And, and then I love to get Peters feedback, but how it got deployed out to all those stores and get it into a synchronized database. It gave me gray hairs, and I wasn't doing all the work. So

 

Rich Alterman  16:05

yeah, and I think, you know, one of the things, Peter, I think it'd be really interesting for you to discuss, as you move from a manual environment into more of a automated environment, I suspect that oftentimes you may get some level of resistance from the people in the stores. So could you kind of share with the team, you know, what was your process for kind of selling the use of this into your business and what was kind of like that migration from going from basically 100% manual to really making that more of an exception versus a rule.

 

Peter Wyman  16:38

So we recognize that there would be substantial pushback from lots of the branch managers who've been around for a long time, and came to realize that while we were reducing their ability to approve loans, based on their own judgment, they would nonetheless be there, their potential bonus, availability at the end of each month would be affected by charter offs, where they didn't have control over the approval process. So there were there were substantial grounds for for some grumbling. To say the least, we actually decided to take this slowly. For a number of reasons, but but a lot of it having to do with that, that resistance, that hesitancy in the field, we started out by basically saying, this is a tool, we'd like to you to look at it as an advisory tool right now. And then we're going to talk to you about how it worked. And you know, how do you feel about it? It's complex enough, they can't mentally reverse engineer it, I don't think anybody can. So really, we started with its advisory. And then we basically said, Alright, this is now guidance, meaning you can still override, but you have to, we had a place in the application where you could document why you overrode it. And now we're at a point where we've got enough data and enough performance information, where we can call out the ones that say, we told you not to make that loan, you made that loan. And here's how that group of loans have performed as compared to all the rest. And it's pretty clear that there's a real mandate for Alright, it's time to, to lock that down. And we've now done that we've now proven to the to the field, that it's reliable, and that it's batting averages is better than probably theirs would be. Right. It's hard for people to accept but

 

Rich Alterman  18:26

so so, you know, when you first start out building the model, there's really kind of two two opportunities, right. One is that you may find there's people that you've been declining, that you should have approved, and people that you approve that you really should have declined. So maybe I'll put this both to you. And Carl, you know, what were some of the outcomes from the models? And how did that, you know, where did they which camps? Did they tend to fall into the approval that they should have declined or the decline that they should have approved?

 

Peter Wyman  18:57

Oh, you want to or over your time?

 

Carl Spilker  19:00

Yeah, I was just going to jump real quick that when we built the model, he ultimately build them on the loans that have been funded by by Brundage over time. So you build the models on that. And we did a bit of reject inferencing and found that there is a sizable group that you could swap in. And it makes sense in a manual underwriting kind of historical play. So when we presented that we tried to boil the ocean a little bit on what's the profile, what's the distribution and make ourselves come out? I'll be honest, that even from the GDS side, just making sure we're absolutely comfortable, we could present this to Peter and their team, and then to articulate that there would be some risk in deploying it, which probably was, I think, wise on the kind of slow rollout. So I'm eager to see if you guys got a fairly sizable swapping opportunity that maybe opened up the eyes at the branch level. I think it certainly

 

Peter Wyman  19:53

did. When when the model was in was finalized. and presented to Brundage executive management. The message that was conveyed is that of the total benefit, net net benefit net of cost about 60 to maybe 65 70% of it was actually loans you didn't make that you should have. And the remaining 30 to 40% were loans you made, you shouldn't have made. And I think those statistics have held up as best as I can see. So yeah, it was it was it was proven true. I think the the real value wasn't just in that that certainly helped us. But there's two other things that it really did. It helped us scale up, it basically allowed us to launch and grow our digital presence. And to give more rapid pre qualification response to online customers, which we could never have done without. And so I think if you said, what made you the most money, that was probably it, right, he's been able to grow the business without adding a huge amount of staff to be able to handle that. And that's a very, very significant win for us.

 

Rich Alterman  21:10

Yes. So Peter, can you kind of talk through, you know, obviously, Carl and his team develop the model, the scorecard model? You also, I'm sure, use a set of policy rules, and maybe other attributes. So maybe, how do you how do you maximize that marriage, right between the score and the attribute slash policy rules?

 

Peter Wyman  21:31

Yeah, the world hasn't stood still has it very much between between 2019. And now, it was this little thing called the pandemic. And one of the things we did we ran, I want to say it was 200 250,000 loan records through the models, first to teach the models and then to test them. And post pandemic, we ran another 200,000 through to see if the pandemic had affected whether the models were still working well. And, frankly, I was surprised pleasantly to see that they were working really was so so that was good news, the places where we have made changes where we have have, I'll say trimmed around the edges is kind of twofold. The models give us a one to 10 score, both for default and fraud risk. And if you look at that as 100 cell matrix, then we color we value those cell values to rate the loan risks in combination of those two factors. Between a these are your best customers and D decline these so ABC are okay, these you don't make. And then with that information and some other attributes, including the borrower's monthly income, we actually calculate line of sight where we say this is probably the best loan to offer them. It's it's high enough that it's profitable and not so high that it has a higher payoff risk. And then we have a maximum number that says no matter what don't get more than this, we have the ability to change those values. With time, we can do that in a state because we have those models at each state, we can actually do it all the way down to the branch level, which we have done in some cases where we had a fraud risk problem in particular. And we've tweaked those over time. So that's something we do without disturbing the model, the models the model that gives us what we need, then we need to interpret that on how to what's the call to action. The other thing that we've done is in the 36% states, which weren't 36% states when we did all that modeling. And so we had no 36% loan, borrower information and performance data with which to train the models. And so what we've done is we've implemented a set of credit overlays that do knock downs and knockouts based on some I'll call it reverse engineering, where basically we're getting credit attributes on those loans. And then we're looking for clusters of risk that weren't necessarily in the models, but appear in some of the other ancillary information that we get either from GDS as attributes, or from the credit bureau alternative data that we get that we didn't ultimately use in the model, but nevertheless collect. And so we were able to look at those and find some pools of risk that we were then able to carve out with some with some knockouts. So we've done a fair bit of that, but particularly in the 36% states.

 

Rich Alterman  24:24

Yeah, so let's since you've raised that let's kind of talk about the the certainly the short term lending industry, the higher risk higher rate. Industry has always been concerned about the potential national rate cap at 36%. Senator Durbin, just recently it came up again, I think two weeks ago, it was in the news. But that said, you have I think 18 or 19 states plus DC that have already implemented a 36% rate cap at the state level. You mentioned Illinois and New Mexico. You know some of our listeners Today may be folks from that short term lending industry. And certainly when states have moved to a 36% rate cap, some stores, or some lenders have taken the approach that we're basically going to shutter, our businesses, and we're moving out. You guys obviously didn't make that decision. So when you think about the profile of that customer, who's who that you're now targeting in that 36%, Illinois in New Mexico, maybe you could talk a little bit about, you know, how you've had to adjust your thinking, you know, different loan amount, how long they're going to be in the product? And what are you finding from a, without getting into any secret sauce, but from a profitability standpoint? Are you feeling good that you made that decision to stay there?

 

Peter Wyman  25:47

Well, we are glad we're still there, we're profitable in both states, we're actually doing very well in New Mexico, Illinois is a little more challenging. But I will say that it did require a lot of adaptation on the part of Salone. For one thing, you're now chasing a different borrower, the borrower's that we have today are not the same borrowers that we had, in the non 36% error, these are basically low prime top high subprime, but somewhere in that middle middle tier, not lower subprime customers, we're giving them larger loans, the typical loan amount, and a 36% stake is somewhere between two and or I'll say 18 104,000. Whereas in a non 36%, state, there are $500, up to maximum, maybe 2000. And you're dealing with a different credit risk customer. So that so the trick was finding the new with the new borrower that we were looking for competing successfully against others who had already entrenched in that market, one main and others that were already there. Obviously, lots of people leaving at the low end of the market really didn't help our case heading because they were having the same problem we were having. So really, it was about making larger loans, consolidating branches, which meant we needed to be again, more efficient, we did a full upgrade of our core platform, which made us a lot of greater efficiency. So in combination, we were able to sort of come up with a new business model that works well in 36% states, and quite frankly, works extremely well in other states as well. So so we're kind of following that that model throughout, probably not as aggressively, we can afford to be in some smaller locations in non 36% states, where we really can't do that as much as the 36% states.

 

Rich Alterman  27:56

Right. Well, thanks. So um, Next, I'd like to talk a little bit about the back end process. So we often see that and I've been, you and I were talking to you, I think you felt like a little younger on the call today, for some reason, but been around here for 40 years now. And oftentimes, when I've been involved with implementing, through software companies, you know, scorecard projects, there's a set of back end reports that they suggest that you need to put in like score, populations stabilities characteristic analysis. So I know that you partnered with Carl GDSL as a model monitoring service. So Carl, maybe you could talk about how you leverage that with Peter, to monitor what's going on with the models. And you know, what's the frequency or process where you may collaborate together and say, Hey, we're seeing some shifts? What are we going to do about it? So, Carl, if you want to pick that up with?

 

Carl Spilker  28:55

Yeah, absolutely. So one of the things about our offer is trying to put myself in St. Peter and his team shoes is not to create a black box. So we're very kind of hand in hand going through the variables, you know, all the different datas getting down to the final model. Then more importantly, as it's being deployed, we're keeping a watchful eye For you know, any any variable drift. So we do kind of a monthly update, walking through, you know, how's the drift? How's the models validating? How are they tracking to the original development set? I think if I'm not missing something, there might have been like a small period where one of the providers data triggered off you, you capture that recognize, fixed it within a couple of days. So it's a it's a good, good catch. It also is helpful because, you know, three to six months as we were just talking about the 36% states, there's you know, adaption that needs to be made and adjustments. So having those conversations then kind of puts is in a position where we can, you know, offer you know retuning type of services where, as part of the program, we're doing a, we've done one retune, I believe and we're doing a full model rebuild for the new loans and moving into the former borrower piece. But but the idea there is you're keeping kind of a pace that is that model going to be consistent? Is it going to predict on exactly what was developed. And if it's not predicting on those levels, it's time to either we raise the hand or management raise the hand obrazy side says, Hey, guys, you need to address this, fix this and puts us in a position to address it, I want to say quicker. I worry sometimes you could get in a six month low before you start finding this out. And by the time that happens, you know, your p&l suffering, right, that that I think very, it's beneficial for us, because we've already invested the data in the knowledge. So it's quicker to make the adjustments. And I think it's beneficial for Brundidge. Because then they know somebody's looking at it. And then secondly, they also know that hey, I know you guys think this is acceptable psi drift, but I think you need to jump on this because you're seeing things that we don't see. So I think that was very, very powerful in that relationship side.

 

Rich Alterman  31:16

Right. Thanks, Carl. So Peter, you mentioned at the beginning that we're certainly in a different place today than we were back in 2018. A lot of headwinds, credit card debt, you know, as broke through a trillion dollars, the student loan federal guarantee pause ended, the first of the month, the average student loan payments can be about 300 to $350. surveys out there that 30 or 40% are saying they may not be able to make their student loan payment without making adjustments to other spending, new car average payment $750. And, of course, the inflationary impact. So all of these things that, you know, certainly are impacting consumers. And maybe this is for both, Carl, you and Peter, you know, have we been seeing some changes in that through the door population, in their ability to, you know, to pay? Is that Is that something that you think you're gonna have to make some good adjustments for?

 

Carl Spilker  32:19

Yeah, I'll jump in. And then Peter, just kind of let you go. Because I remember there was that anxiety during the COVID period. And that was, me personally, I was less worried about that. I'm more worried about this anxiety during this transition period. And there was a bunch of work to make sure we, we built a pre COVID It's still work post COVID. But what I believe we're seeing this, but I'm seeing across the board from many clients that we work with, and we don't work with, there is a rising level of delinquency. And I believe that just means we've got to be in a space to basically retune redevelop, identify shifts very quickly. Because we like to articulate as a curve shifting upwards low risk, people are getting higher risk, moderate risk are getting moderately high risk, and high risk are triggering to higher higher risk, it's happening across the distribution, which means the money that's being deployed into low risk and higher limits is going to absorb a higher risk. So that's that's going to put you know Brundage into place to adapt for that. And policy helps them adapt quickly. And then redevelopment helps us refine that tool.

 

Peter Wyman  33:34

Here. I'm going to step back to last year, particularly the second half of last year, where a lot of lenders, I think, started seeing the first bits of some of this, it was after the COVID stimulus money had dried up. And people were having to go back to, you know, living on their their wages. And it was it was challenging time. And I think you saw a lot of letters, including us adjust some of their credit limit and some of their underwriting practices at that point. And so the challenge is don't don't pull back so much that you're not you're not maintaining your ledger, I'm pleased to say that what we've been able to do is manage our delinquencies and bad debts. And nonetheless, gradually grow the the ledger balance will actually be we think at the end of this year, we'll be at an all time high in terms of our in terms of our book of business. So we're pleased about that. But it was a case of don't press too hard on the accelerator, take it, take it slow. Make sure you're getting the right customers and that you're maintaining relationships with those customers. And I think that really does make a big difference. So we've been we've been very careful and thorough in that. And I think again, that may be one of the advantages of having a large branch network is you have an opportunity to lay eyes on your customers and to develop those relationships. Where if you're solely digital, you really don't.

 

Rich Alterman  35:12

And that is a question I forgot to ask you. Do you have a digital channel? Oh, yes. Oh, are those loans done 100% online? Or do they have to go into the store at some point,

 

Peter Wyman  35:24

they do not have to go into the store. But we will be getting their borrower exhibits certain of their information electronically, we certainly give them the opportunity to go into the store, but they're not required to do so.

 

Rich Alterman  35:37

And, Carl, is there any any model separation based on Channel? Maybe that's a question Peter get answered? Of course.

 

Peter Wyman  35:46

No. And the reason was, most of the data that we were the vintage that he was getting, we didn't really have much of a digital channel to speak of anyway. So we had we had more of a contact me webform rather than a full application. And we implemented all of that stuff, really, in the latter parts of the data, vintages that we were examining. And so so the answer is no, that wasn't something we were able to do.

 

Rich Alterman  36:17

Okay, good. Good. Well, thanks for that. So so let's talk about really that back end process with whether it's account management or collections. So you've embraced analytics on the front end, there's certainly opportunities to leverage analytics. On the back end, maybe you could just talk about some of the things that you may be thinking about, on how you can pull analytics into that side of the business,

 

Peter Wyman  36:42

we get a wealth of information, particularly payment information from our existing customers that we can then use to look for patterns of behavior relative to their payments, those actually do factor into the former borrower lending model that Carl and his team built. But they're, they're pretty useful for collection activity as well. There's, there's some, there's some ways of highlighting stuff statistically, that helps you understand which customers you need to reach out to first. So we're we're focused on that we're focused on an account management strategy that we will implement next year, that basically tries to track a customer through being a new loan customer to being I'll call it a more seasoned or Premier customer. And so how you underwrite and how often you will let them renew loans and things like that, at what limits you will impose evolves over time as that customer graduates from being a basic customer, to a premier customer. And so those are things that we that we're building internally right now that we think will will help us a lot. The other thing that we're doing, I think we're actually behind the times on this one, but we're going to be not selling but basically assigning charged off loans for third party collection activities. And we believe that will, will yield us some benefit as well. So great, great. Yeah.

 

Rich Alterman  38:09

Cool. So I might go back a little bit cuz there's a question I forgot to ask. And Carl mentioned the attributes. He was specifically referring to credit bureau attributes. GDS also offers a set of banking attributes where we can take in data through plaid or Yodlee, you went from a manual process? As part of that, why are you guys picking up bank statements? And if so, as you move to more automation? Is the use of services like a plant or a Yoli? Something that you've also incorporated?

 

Peter Wyman  38:40

We have not yet. I think, when you, when you look at some of the costs associated with that data, we had concluded to date that we weren't ready to do that. I think as we start going into higher dollar loans, when we get into, you know, $10,000 loans, I think that's a place where we're going to want to look at people's cash flow more carefully. And I think that's the point when we'll jump into a plan or Yoda, Lee and whatever additional attributes that we need to be able to evaluate that the data costs are a source of concern. Also, you know, to be honest, we haven't been at this that long, it's been two years. So so we're still I think, you know, learning how to make best use of what we have, right? And I want to make sure that we do that before we go off looking to spend money on something we don't know how to actually take benefit from.

 

Rich Alterman  39:39

But if you don't mind, I'm gonna let let Carl give a quick little sales pitch if he may, on some of the analyses we've done and how open banking data can add some lift. So Carla, a little selfish plug there if you want to take a minute.

 

Carl Spilker  39:53

Yeah, we did chat about this in the last Executive Session, but we have been building models is leveraging the the open banking data and been able to generate some really strong lifts even for Prime and subprime clients. It's interesting, I kind of look at credit data as a little legacy drag on how well you paid or not paid. And then the cash flow is the more recent indicators of how things are performing. It's a little bit of the one plus one equals three equation. And it really helps, you know, for bigger loan amounts, and then it also helps for a really heavily homogenous delinquent group to tease out that additional information. But But Peters, right, the permission piece adds that bit of friction. Many cases, the KOSPI can be a bit of a painful thing to absorb into the flow, but but the market is changing, I think costs are going down frictions becoming more acceptance. So. But we've got three plus years of experience around that as well. So that Thanks, Rich,

 

Rich Alterman  40:59

you're welcome. To send me the check. Oh, good. This has been really interesting, just maybe to close before we get into some q&a, going really, to the very front end, and that would be marketing. Just a one of the things that might be interesting, Peter, as you look at the states, where you're at that 36% rate cap versus the other five, you know, as your marketing department had have, have has a marketing department had to adjust its messaging. When you think about attracting that consumer in those in Illinois and New Mexico versus the other states,

 

Peter Wyman  41:35

I think the marketing department spends a good deal of time focusing on specific customer groups and segments. I think the the the focus on 36 versus non 36 in the borrower profile is really probably handled through some of the targeting that they do with propensity models and things like that. So I'm probably not the best person to answer but I suspect the answer is yes. But it's, I'm not sure how sophisticated that is, I suspect it's probably some zip code and also propensity related.

 

Rich Alterman  42:12

Great, thank you. But I think I think we'll we'll open it up now for some q&a. I guess peope will lean on you a little bit to see if we've been getting questions along the way that we can present to the group and see how that goes.

 

Pia Wilson  42:28

Thank you Rich. Yeah. So we do have a couple of questions in in the chat. And just remember, folks, if you want to, you can just click on that chat button and type your question in there we can I can read it out loud. And you or you can also direct it directly to either rich Carl or Peter. But I'll read the first one here, it looks like Peter, this is to you that says given your state by state model, how do you manage variations in regulations and lending practices across different states? And how do the custom models can adapt to these variations?

 

Peter Wyman  43:06

So the business rules in our origination system are crafted to adhere rigorously to the state requirements at each state. Those in turn, have been I'll call it translated back into the line assignment and the approval grids. So it's, perhaps it's a little counterintuitive, but we started with what are the state regulations? What do we need to put in our core platform? How do we then express those limits into the underwriting modeling?

 

Pia Wilson  43:42

Great. Thanks, Peter. Looks like we have another new question. Do you use the platform for any pre qualified credit marketing? If so does that require a second path for decisioning?

 

Peter Wyman  43:58

So we're in the process of moving to a soft poll for pre qualification through our existing GDS models. And what we will then do is send a file to the credit bureaus for those customers who actually funded a loan with us. So that's being done. As we speak. I suspect it will be done next month. We're not using the GDS models today, in any way related to prospect, mail marketing, mailing or things like that. Because the way we've set our ours up, it's a transaction by transaction approach. And the data costs are very high relative to a prospect that you have such a low response rate that the cost gets out of hand.

 

Pia Wilson  44:57

Right. Thank thanks again, Peter. So, Peter or GDS team, here's another question for you. Are there any lessons learned or maybe unexpected discoveries during this project that you'd like to share? Which might be valuable to other organizations considering similar initiatives?

 

Peter Wyman  45:19

Carl, you want to start with that one? Yeah, the, for me,

 

Carl Spilker  45:24

the biggest eye opener was level setting all the various data sources out there, and letting them compete on an empirical sense. So I know we had a probably a fun, three to six week when we're evaluating these things, a bit of the eye openers, you know, things that you know, you're on one hand is the greatest thing, since sliced bread doesn't really play in, and then how to intermingles i, we could probably spend an hour just poring through what what we uncovered there, that was really good, because like I said, we, I forget how many how much data that was we went through, but the findings were really insightful. And then to play it out to a cost. Because there's, there's some providers are really proud of their data. And it's so costly that to get a little lift, and AUC doesn't pay for it. So that was, to me quite eye opening.

 

Peter Wyman  46:23

My part, I felt like I was going to finishing school, I learned a lot. One of the things we haven't mentioned that Carl's team did was they figured out that when you had borrowers that took out a loan, and then renewed several times, the intermediate renewals didn't really matter what mattered was the initial credit decision, and the ultimate low performance. And so we created loan chains. Out of the information, that was a breakthrough idea. I don't know if it was new for them, but it was sure I'd never thought of it. I never thought of our business in that way. And it made a made a big difference for us. So I think that was huge. Certainly the comments about the diversity of data quality, versus cost was interesting to say the least. And then I was expecting the human impact of this in the field to be challenging. I ultimately looking back on it, I feel like we did a good job. One of the things that we did takeaways is we build a feedback report that we present to the user of the system that gives them not only here's the information you gave me, and here are the scores. And here are the reasons why it's not better and all that. But basically as much transparency, and a very readable format, that really helps the user be a little less wary about what we're what we're doing. So I think that helped a lot.

 

Carl Spilker  47:48

I would agree with that. Peter, I was expecting a lot more gnashing and screaming and things of that nature. And I think you guys set the expectations of the field where it matters, because that's where the rubber hits the road. That was impressive on your part.

 

Peter Wyman  48:03

Thank you. Thank you.

 

Pia Wilson  48:05

All right. So we did just get another question in, can you touch on the high and low loan limits that are presented to the customer? Follow up to that, does it provide any confusion? Or does the model determine the interest rate or credit tier.

 

Peter Wyman  48:23

So the the model, as it works today, the model basically says, irrespective of how much money the borrower requested, here's the best amount to offer. And here's the maximum amount to offer based on the credit information and running through all the models, and also taking into consideration the borrower's monthly income because again, we're we're in the business of making sure that we're making loans to people that can afford to pay us back, right. So it's kind of kind of nice. If you take that into account, what we do, our origination system uses static loan amounts. In other words, you don't define the loan amount as anything you want between x and y. There's, there's individual loans. So we find a product that fits the numbers that we just gave them and say here are the loans you should offer. And then if the customer says, Well, I really didn't need that much. I just need this. That's fine. That's good. We're not We're not telling them to borrow more than they should. But we're telling our we're telling our field don't offer more than this. And this is probably the sweet spot for that customer.

 

Pia Wilson  49:30

Great. Thanks, Peter for answering that. We've got one more. I don't know if this is for GDS, or Peter. Either of you can take this How long did the model Build take with all of the sources internal and external?

 

Peter Wyman  49:45

I can answer that. We got started in I want to say it was the end of January early February of 2020. And we went live For the following year in May, part of the reason that it took that long was because we spent six months building the models, again, I want to emphasize, we did not have a process that they could emulate. We had a blank piece of paper and said, Oh, by the way, crop, can you please fill this in. So, so it was a, it was a greenfield opportunity. And we really approached it that way. So it took a long, it took a long time. And then the remaining time was spent, basically constructing the mechanisms to run the models, and to return all of the information. And by that I mean, all of the information. So we can do post processing analysis of things. And we went live where we tested for a good period of time, and then went live starting in May, and rolled it out very slowly over the next three months or so. And that was May of 2021.

 

Pia Wilson  50:55

Right. Carl, did you want to add anything from your team's perspective on kind of the the time and and or typical?

 

Carl Spilker  51:02

No, no, what I was going to add to that was the discipline of an executed program and plan, I think is absolutely critical. Because not only we we sell and Peter and his immediate team, but Peters got to resell that through the organization. And that was very patient. And notwithstanding, they still hold us to a high standard. We know the model deployed, but make sure it's still validating the tracking as we developed it was was good because it meant we felt comfortable that we're delivering a good product that will then when it gets deployed, it'll still mean what it meant. So yeah, I liked that approach is very consistent, very disciplined. And I hope it's paying dividends.

 

Pia Wilson  51:48

Right. Looks like we got one more here looking back at the project. Are you considering the possibility of implementing a similar initiative in the future? If so, or? Or what initiatives? I guess, are you considering? What are what changes would you make?

 

Peter Wyman  52:06

Well, as we already mentioned, we're refitting and and reviewing and updating all three of the models that we currently have. And it's just time the models have survived the test of time quite well, considering everything that went on in between. But it was time that it was time to do some some adjustments and corrections. I suspect that will help I'm I am still interested in the whole area of banking information, whether credentialed or non credentialed. And I'm I'm certain that we'll be doing something in that space. Eventually, again, particularly as our loan amounts are continued to increase. So it's a case of not if but when I would suspect those are probably a couple of the things that we'll do. There are some other areas that we're poking around with. The the account management, although it's not a formal model, is something that we're doing basically, with with data attributes that we collect from our servicing system. And I think that will help us as well, we also obviously have all the credit data on that borrower when they took out their loans. So we have, we have a good body of data. And I think we'll be building some, some some not terribly sophisticated models, but nonetheless, some some some analyses that we'll use to drive that process. So those are probably the areas that I'm that I can see in the in the short run over the next, you know, 12 months or so.

 

Pia Wilson  53:35

Right. Thanks for that insight, Peter. We're looking through I believe we've answered all all questions. Again, feel free to reach out to us individually if if anything else comes up. Again, thank you rich Carl and Peter for today's informative session. But before we wrap up, we have three important GDS link updates to share with you. First, if you're attending lend 360 Mark your calendars for day two at 9:45am. We have an engaging panel discussion titled unlocking the potential of emerging populations with advanced risk analytics. This discussion will feature our very own rich Ultraman and Josh Martin next from Brundage seconds if you haven't already done so we invite you to please explore GDS links podcast the lending link, you can access it at GDS link.com/the lending link, or it's available on all major podcast platforms, so be sure to check out our latest episodes and subscribe today. And our third, we are thrilled to announce that we will be hosting another live webinar to introduce you to GDS links groundbait breaking widget our E pre screen 365 This webinar is scheduled for now Next Thursday, September 28, also at 1pm Central. If you have any questions, please don't hesitate to reach out and keep an eye out for a follow up email, which will include access links to all three of the items I just mentioned. And thank you again for your participation. We will be sending out your Grub Hub gift cards via email in the next few days. So make sure you keep an eye out for that. And lastly, if you're interested in hosting an upcoming client Connect event with us, please feel free to reach out to me pa dot Wilson at GDS link.com. And thanks again everyone and looking forward to connecting with you all soon. Take care. Thank you