The $10B Insight Every Lender Needs

0:00 It allows us to move from just predicting who will default to actually shaping the default. We're managing more than ten billion dollars in loans today, and while that might seem paltry compared to maybe some of the big five banks, of course it's a really, really large portfolio. If you think about all of those barbers who are delinquent, and you take the ones who are definitely going to pay you back whether or not you talk to them, well then you definitely shouldn't talk to them. Lms are, of course transformative. They're an unbelievable technology layer on our work, and I sort of imagine that they will eventually be diffused into almost all. Of our activities.

0:43 Welcome back to Leaders Lending, where we explore what's next in the worlds of lending and financial innovation. I'm your host, Lynd Saderbil, and today we're talking about a part of lending that doesn't often get the spotlight. Servicing. It's where technology, operations and customer empathy all meet, and at ups art it's becoming a powerful driver of performance and borrowers success. Joining me today is Jesse Opink, who's the general manager of Servicing at Upstart. Jesse, do you want to introduce yourself for our listeners. Ellen, thanks for having me excited to be here. I joined Upstart a little bit over seven years ago, and. I came up on the product side. I was a product manager number two are our product management function was relatively late to develop, you know, and I've gotten a chance to work across our personal owns function, a little bit in growth, product, a little bit with machine learning, and and now for for the last two years, I've been the general manager of servicing, you know, which I think really combines my passion for great customer experiences, you know, with with my desire to run scaled in and highly effective businesses.

1:55 Awesome. Well, I think the one thing I think, you know, we we do a lot of unique things here at Upstart, and we do I think a lot of them very well. And this is an area that's been a big area of focus for the past couple of years as you've become the general manager and have been really leading the servicing, product, engineering and operations teams. One of the things you've said before, and I think you may have shared this at the client summit with many of our clients and potential clients, is that servicing is really just underwriting with an operational twist. What did you mean by that? Yeah, you know, as you as you mentioned, I have been at Upstart for I guess seven years last Saturday, so just just past the seven year mark. And of course I had spent the first five years working, you know, very deeply in upstarts core business that is predicting who will default and then pricing them correctly, our our underwriting business.

2:57 And so when I joined servicing a little over two years ago, I really tried to think quite a lot about what it meant for us to take our core competency and apply those skills to this relatively new space. And I guess the way that I would say is that in underwriting, your job is to take maybe all of these applicants. You're taking applicants, you're sort of lining them up in riskiness, you're spacing them not correctly, and then you're determining generally what risk based price is associated with the level of risk that you see for each applicant. And I think in many ways you really want to do the same kind of thing with servicing. You want to maybe take all of these loans, you want to line them up from least risky to most risky. But in underwriting you have this sort of unfair advantage that is that you can maybe take any person who's risk is too high, and if you don't like how risky they are, well you just don't offer them alone. And in servicing we actually have a little bit more operational complexity.

4:01 We are forced to maintain our relationships with these customers, and so instead of being able to maybe just say, oh, this person is too risky and we're not going to work with them at all, we instead have to maybe look only at the range of operational levels that we have to change their riskiness today, to move from how risky they are today to how risky they might be tomorrow. And I think of this as a really really powerful part of our overall opportunity at Upstart, because it allows us to move from just predicting whole default to actually shaping little defaults. That's a little bit what I mean. Yeah, Now it's super I think super interesting is really kind of applying the the data science or behavioral predictions after origination, I'm kind of dynamically handling that relationship as you move forward with the customer. I like the idea of what you said of You know, if you identify a higher risk pre origination, you can than their APR.

5:01 If you identify that after you have to still figure out a way to work with that customer and to maximize the returns that you're getting from them. That's something that we've been actually really focused on in Servicing. That is like, oh, we you know, not only want to be good at the risk prediction part of our work, but we also want to have a really wide range of operational levers. And so one way tactically that this has played out for us is that we really want to have collectors with a range of experience and persuasive capabilities. You know, you might imagine that maybe on the extreme, if you are collecting on very very small dollar amounts, it's it's probably actually not worth it for you to have someone really high end making that call. You might actually even want to stick with just digital only collections.

5:54 But on the other hand, if you have someone who maybe owes you tens of thousands of dollars, maybe has a a secured asset that they really care about, you know, and and maybe need some coaching to get to a good a good outcome, well, that starts to look a whole lot more like artisanal sales. It looks like a really persuasive and engaged conversation, you know. And and so uh, I guess our work in the last couple of years has been both trying to build out our risk predictive capabilities, the ability to decide which loans are risky and which are not, but also a much broader range of operational capabilities from collections, hardships, settlements, uh, you know, and uh and recovery options.

6:36 No, that's I think that's super interesting. And you know, and certainly the business that you're running has a really large scale, uh that it is a large business that we are servicing and across multiple products now versus just the the original personal loan uh fixed rate installment loan product that Upstart is known for. So maybe talk to me a little bit about like why is the servicing operation different, particularly for the size of the organization. Yeah, I definitely do think there are some parts of servicing that are really well plumbed, you know, and well known in the industry, you know, and then some parts that really give us some opportunity. And maybe I think of us generally as we scale as having the opportunity to do more and more things with novelty. That is, like, you know, we're managing more than ten billion dollars in loans today. And while that might seem paltry compared to maybe some of the big five banks, you know, of course, it's a it's a really really large portfolio, you know when you think about it at an absolute level, and that sort of gives us the opportunity to really make deep investments in our technology infrastructure. You know, we have a large engineering team. We have not that large, but very high talent and growing machine learning team, you know, and we have a set of vendor partnerships you know, in the in the low low dozens, and these are

8:09 these are kinds of infrastructure and partnerships that you know, most lenders just can't justify on their own. If you think about, you know, maybe trying to set up this kind of a technology organization. You know, if you were managing a portfolio of maybe uh, you know, one tenth or one fifth our size, then trying to amortize those costs wouldn't make sense. And you know, we're really quite excited about the opportunity to build out that technology in a way that deliberately plans for scale and then use the sort of operational leverage of our scale to reinvest in more technology build out. Yeah. Now, I think that's a really interesting point, Like you think about the things that upstart is known for, and we do really well our core competencies, and then applying it to this area and really being able to do kind of the basic things that content acts, compliance, collections, all the specialty items you mentioned, but then really having that enterprise level infrastructure run like supporting it and learning from all of those repayment events that your team sees every day.

9:13 To bring it back to that first point that I was sharing a little bit earlier, I think this is a perfect machine learning problem. That is, it's very high analytical complexity, you know. That is, Oh, what we care about in servicing is not really whether someone pays us just tomorrow, but whether they pay us over the next three, five or seven years depending on the loan term, you know, and deciding whether we should be really pursuing collections hard, whether we should be approaching someone with an offer of a modified payment schedule for short term relief or actually offering a much longer term and more permissive settlement. Well, well, this is this is something that has an objectively correct answer but is extremely extremely hard to predict. So the combination of high analytical complexity, massive personalization, and as we were just talking about this sort of ten years of data from Upstart's history and servicing, I think this is this is something that really gives us a great position to build out the kind of scaled program that it would just not be economical for smaller portfolios to deliver.

10:21 Yeah. Absolutely, I think the scale is important. And you know, it's a lot of the reason that that partners work with us on the on the underwriting and customer acquisition side, right, Like, we have the very large marketing engine to to identify and acquire the customers that they will want to convert into members or or into customers and cross sell them, and we can apply that same scale in this area and really deliver a delightful experience while maximizing their returns. One thing that you said that that kind of caught my attention is that you've got in a third party's and vendors maybe in like the dozens that you're working with, and so you know that is certainly an operational challenge, and how do you think about all of that vendor management and like quality and oversight and more of kind of the core operational function of working with those sorts of BPOs and vendors.

11:14 I think this is such a good question because there's really a strong tension in servicing and that is maybe number one that I want to invest our energy in doing things that we can do differently and better than other kinds of organizations. You know, and boy call centers, they've been around for a long time, They've been optimized in really really deep ways. And at the same time, you know, we really want to set a very very high quality bar and we want to make sure we're delivering better than simply industry standard. And the way that we think about that in relationship to vendors is really that it starts with pretty clear accountability. We have this dual ownership model where both our internal teams and our partners really share responsibility for these outcomes. And the way that we sort of materialize that in the world. While we run random audits, of course, we try to spend a lot of time with our vendors.

12:08 We're doing a lot of site visits. Uh, you know, we have a pretty clear performance scorecards, you know, and then we want to make sure that actually those things are backed up with very clear quality incentives in our contracts with those vendors. You know, we want to make sure that our incentives are aligned and when we do well, they do well, you know. And and generally what that means for us is like, if, for example, an error rate it creeps above our target, then we want to make sure that there are contractual penalties. But if conversely, quality improofs the partner is going to win more volume, they're going to be able to grow, and they're gonna be able to sort of benefit from that relationship with us. So you know, in this regard, you know, I would say that managing vendors well and doing doing contracts, this isn't really the flashy part of servicing.

12:56 It's really just about building trust and enforcing consistency and and really identifying root causes really clearly when things go wrong. Yeah, and I think so less scenario where you'd maybe focus on novelty and more about just being precise and having discipline and having good communication between you and the vendors as well to make sure that you're all I think agreeing on what quality is and how you're measuring it is certainly critically important. I love that you use that word novelty. There's a quotation from Angela Duckworth, who I think is one of the foremost thinkers on grit that talks about how you maintain focus in an area when it becomes a little less fun And I just remember so clearly her sort of saying, Oh, your ability to substitute nuance for novelty, to think about doing the same thing but with ever increasing nuance is highly predictive of your willingness to stick with something, you know, even once it's no longer flashy, you know. And I really think this applies to servicings so deeply that the best servicing models they're not going to be they're just really relentlessly consistent.

14:07 Yeah, that's funny. That's one of my favorite quotes, and I'm probably going to paraphrase it a little bit wrong. Is that like a differentiator between people as people who are willing to sprint when the finish line is unclear, but are willing to sprint anyways. I'm going to take that. I love it. So, you know, we've talked a lot about like the machine learning pieces, and then some of the core competencies and the vendors and different things that you've been working on with the team that maybe talk to me a little bit about like the people side. So when you think about like identifying that differentiated talent, like, as you said, you know, if you're collecting on say a fifty thousand dollars loan that's passed you, you may need a more skilled senior collector. You may be spending a little bit less effort on something that's maybe a smaller dollar or maybe less likely that you don't need is skilled of a person in there. So you know, maybe kind of a couple of things differentiating talent, like identify people who are the right fit for the different roles and what those roles are. And then as you think about all of the work with LLLMS and automation, how you kind of bring humans into the loop and where the humans are there, and then where the automation is best the kind of forward forward point.

15:18 I think these are both really really rich topics. Maybe we can start with sort of thinking about human talent and then talk about how we're incorporating ELMS, because of course that is the technology shift of our age. On the human side, I guess I really tend to think about the first and most important thing as being able to understand what causal role particular human is playing in a servicing decision. And what I mean by that is like, if you think about all of those barbers who are delinquent, and you take the ones who are definitely going to pay you back whether or not you talk. To them, well, then you definitely shouldn't talk to them.

16:00 You know. There's no there's no reason to spend your time and money trying to reach out to somebody who is sort of gonna pay you no matter what. In that case, the human would have no causal impact. And maybe on the way other end of the extreme, if we take people who are definitely not going to pay you back whether or not you call them, well then don't call them. You're wasting their time and you're wasting your own money, you know, and again you're having really no causal impact. And so the first question for us has really been trying to isolate the causal impact of particular individuals in changing people's mind I often talk about the best collectors as being highly empathetic, highly persuasive, you know, and there's a there's a big analytical question in sort of how we attribute a certain amount of the collected dollar value to the exact person who was on the phone at a given time.

16:55 This is one really really interesting question. But even once you have solved that pretty rich analytical problem, and we have not solved it, we're sort of making advances, but it's not all the way solved by any means. Even once you've solved that really rich analytical problem, well, then you can sort of think about, oh, well, are the kind of people who demonstrate the greatest causal impact. What kind of background do they have? You know, do we want to actually hire for that background? Do we want to train for it? You know? What are we actually willing to pay? You know, of course you can get generally a different kind of person if you're willing to pay a lot more. And you know, I aspire for us to maybe have some collections roles that are highly competitive with inside sales. You know, I think that generally, if you think about the profile of a collector, we really want that profile to look more like an inside salesperson and less like you know, a call center operator, you know, and then the third thing really how are you incentivizing them?

17:58 What actions are you encouraging them to take? And and what are you rewarding them for and and so I think the whole structure on the human side is really really nuanced and complex in terms of combining, uh, the background of your collectors, the overall compensation, and then sort of how you mix based comp variable comp with specific collections based incentives. If we shift to the l M side of things, l lms are, are, of course, uh transformative, They're They're an unbelievable technology layer on on our work and I sort of imagine that they will eventually be diffused into almost all of our activities. That that is, like, we want them to be part of our software, We want them to be part of our internal tooling, you know, and of course we want to be part of our operational interactions, uh with our operational interactions.

18:51 Uh with borrowers, you know. And the way that that's manifested today, well, we generally have already rolled out programs that allow for our own frontline agents to benefit from better M assisted tooling. We have LMS analyzing, analyzing callscripts, maybe helping to identify where there are complaints, helping to categorize where we have opportunities. You know, we have lms assisting in our training, trying to you know, role play with some of our incoming operational agents, and of course, as with others, you know, we really have started to put lms in a place where they're taking some of these ropes or routine high volume tasks like summarizing notes, triaging inbounded messages, or guiding income in some cases next best actions. And our frontline team has generally responded really really positively to this. That is, like, if you are a skilled and influential collector, you don't really want to be taking notes all day. That's not the part of your job that you enjoy, you know.

19:59 And so some of the feedback that we've gotten from our team is that the system generally feels more organized, feels more understandable, feel more clear, you know. And our goal here is for AI to play a pretty central role in helping to surface relevant context instantly, which we think will let our best people focus on empathy, on negotiation, as I said, on persuasion, you know, and exercise that that human judgment, which we think is is still pretty pretty unique. How do you think about maybe then finding that talent so as you think, you know that is the role is changing. And I talked a lot about and I think it's a pretty uh maybe a unique idea that you want to think about. Uh, people who you know maybe are are currently paid more like contact center specialists, and in the future it could be paid and compensated more like an inside sale like a BDR inside sales representative.

20:53 How do you think about identifying or and or upskilling people that are in these roles today, so as you think about helping them get comfortable with with AI l ll ms and and have the right skills and then and and that I think that mindset of that this is an enhancement, this will make my job better, not like this will just replace my job. Uh. And you know, focus on the opportunity and out the fear, Like how do you think about like identifying that talent in the existing pool, but then also you know, as you're as you're hiring and looking for that outside with with new roles. Yeah. Uh, it's it's such a good question because of course, uh, you know, if you're going to go out and and you sort of say, I bet there are a bunch of people who would be really good at this job, who would never think of themselves as good at this job.

21:40 Uh. Well, that's it's a really really hard search problem. And I guess the way that we've thought about it is by really creating a virtuous cycle, a little bit of a flywheel. That is, Oh, first we put in place this measurement, trying to see how, uh you know, how each person is contributing to the overall impact of the team, and then maybe looking back at those people who are most impactful and trying to isolate which skills or behaviors they demonstrate that maybe other folks don't demonstrate, and then of course taking that both into our training and into and into our hiring, you know, so holding a higher bar for those skills and behaviors, you know, generally rewarding those skills and behaviors. And of course, as I mentioned, all of this is really tied into compensations. So if you're looking for either higher quality expression of the same skills and abilities or novel skills and abilities that maybe someone has and another person doesn't have, well you have to be willing to pay for that. And generally, my hope is that our collections team over time is better and better compensated because they are so good at helping barbers find paths back to current they're playing a really pivotal role in shaping people's minds, their behaviors, and ultimately the repayment on their loans.

23:04 Yeah, and I think it's a unique way that we've talked about this in UH in in servicing here at Upstart, where you know, a lot of times folks can think about operations teams as more of a cost center and it's just you know, preventative or defensive in servicing UH. And I think the way that you've approached it and are thinking about it is more dynamic than that that this is a place where where there's returns to be to be had because there are people, as you said, who there's some people that will that will pay if we contact them, or will pay if we don't contact them, and then there are people who do the opposite, like we can contact them and they won't. But there's you know, I think you have a pretty strong belief that there's a subset of folks that maybe won't pay if we don't contact them in the right way or the right frequency.

23:51 But our behaviors can differentiate that and can drive a better return. You know, maybe talk to me a little bit more. You know, we've already talked quite a bit about machine learning and service and kind of being a hard problem, and I don't think it's an area that really anyone else is spending a lot of time focusing on and using machine learning to solve some of these servicing challenges. Maybe just expound a little bit on like why is it so hard? And why is it so hard to take something that that we know how to do really well machine learning and apply it to this particular kind of set of problems. Yeah, it definitely is hard. And you know, I think there are increasingly folks who have awareness that that servicing is an exciting opportunity space.

24:36 But you know, much as I hate to say it, when I was a little boy, I definitely did not grow up saying, oh, what I want to do in my life is run a large servicing organization. That that just isn't It's not it's not super. Common, you know, And and I think if you you know, if you think about the space and the relevance for machine learning, well, the servicing data, it's it's fundamentally, there's fundamentally a lot of it that's really good, you know, that's that's one sort of core premise for being a good machine learning problem. But the servicing data is it's really messy and it's really behavioral. That is, you're dealing with really complex interactions. Somebody can pay this month, not pay next month. You're dealing with multi channel interactions. Hardly anybody wants to be the lender that is doing a holdout where you're like, well, I wonder how impactful calling is, and so I'm not going to call fifty percent of my barbers and I'm gonna see which one's default. That's that's really not what anybody wants to be doing. And so trying to sort of inch your way toward understanding causality across payments, calls, hardship plans, you know, third party collections, first party collections. You know, the sort of evolution of that data space over time is really really quite complex, and in many ways, you know, it feels to me like we are inching toward an understanding, and every time we take a little bit of a step forward, well, it broadens our horizon to the future. We have a

26:08 little bit of a sense, but there's still so much that we don't know that we can't quite see too yet. Now. I think like if we look at machine learning either, and that's sort of that's a servicing view. If we look at machine learning, traditional mL models, well they're mostly built to predict relatively static outcomes. You know. They want to predict something at a point in time. And the challenge for us is like, well, we're trying to build models that then adapt in relatively real time and predict a wide range of hypotheticals. That is, like which intervention today has the greatest chance of improving recovery or improving customer satisfaction tomorrow. And this is an enormous modeling challenge, but it's also, I think, really a natural fit for what Upstart is good at, and a natural extension of the skill set that we've been building.

26:58 We spend a decade really. Trying to get these models to be good at understanding complex behavioral data over long term outcomes, you know, and we think that incorporating some of these operational feedback loops is a is a really really exciting next challenge. Yeah, and I think it definitely goes into you know, kind of what we started on as we were talking about underwriting and origination and some of the things that impact folks on that end, like the macroeconomic cycle, things like the you know, a recent government shutdown of course, and certain people being furloughed and not receiving their money. You may have borrowers who had a desire to pay but not an ability to pay, which is very different than somebody who doesn't have a willingness to pay, or it doesn't have either a willingness or or an ability to pay.

27:46 At the end of the day, servicing is oh, I should say, loan performance is really a function of your underwriting, your macro and your servicing. And of course, you know, we want to play a central role in upstarts overall long performance. But you're absolutely right to highlight that there are such complex and confounding factors that understanding which part of that impact came from us is a really ongoing challenge. All right. So we've talked a lot about what you've been doing, you know, and I mean, I think it's fair to say that most people don't as you said, is it you know, six year old and aren't thinking I'm going to run a large servicing organization when I grow up, and I would imagine that six years ago you probably were not thinking that either before you you became the general manager of servicing here at Upstart. But certainly looking ahead, I know one of the big topics on the horizon for us as servicing as a service. So maybe talk about that, like what does that mean, and what does that look like, and what would I would kind of excite you about it, and what do you feel like you really need to focus on to be very really effective there.

28:56 Yeah, maybe one of the first things is just to say I really think of good servicing as good stewardship of our barbers. That is, like, oh, I believe that by offering customers better experiences, you make them more loyal to you. I believe that by offering lower friction experiences, you make them more likely to pay you back, you know. And I believe that by offering empathetic hardship programs that match customers real life financial circumstances, you can really help them get back on their feet and keep a loan current that otherwise might have rolled to charge off. And so we've talked a lot about the analytical challenges of servicing, But at the end of the day, the customer experience is really front and center. The way that you create better loan performance outcomes is, in my opinion, by offering really next level and excellent customer experiences.

29:55 And when I think about the magnitude of. This challenge, you know, almost within I don't know, within probably two months of joining the servicing team, I was pretty clear this is a year's long, you know, maybe decades long kind of problem for us to solve. This is not a fast problem. And my goodness, if we build the infrastructure to solve this problem, we shouldn't keep it to ourselves, you know. And that's really where servicing as a service comes in for me. You know. I believe that the infrastructure, the kinds of analytical approaches that we're building, you know, and the kinds of experiences that we're trying to to marry them with are going to be unique and gonna be really best in class.

30:40 And I want to make sure that we're able to apply those excellent experiences to as many loans as possible, whether or not they were underwritten by Upstart or not. Yeah, I think that's a really exciting, you know, exciting idea as we think about you know, kind of optimizing what's working here and then as you said, share it and like we you know, we there's a lot of things I think we are have very strong competencies in and and this being one of them, and so kind of building that for for what's next, and and potentially servicing you know, loans that didn't even originate on our platform as an exciting place for us to head uh, and that we can really bring all of that kind of machine learning uh and servicing know how and applying it to to in a kind of a long term way, which is I think interesting that this is a long term innovation build. So I think what what are you most excited about? Like what's next on your agenda? That's probably to you the most exciting thing, uh, that that you'll get the chance to build.

31:41 Yeah, our north star here is really to be able to service not just UPSTARTE originated loans, but partner loans as well. And we have a handful of things that that we still need to work on our core platform capabilities today. You know, of course our bread and butter is unsecured personal loans. We have lots of work to do to maybe UH build out some of those UH revolving credit capabilities UH for loan types like heylocks or credit cards, you know, both of which make up a very large part of the third party servicing UH market today. You know. But but to be honest, UH, you know, platform capabilities are something that we have to do, you know. And UH, And then that's the part that really gets me excited that I'm that I'm super thrilled about is really trying to drive more of this unique value through more personalization.

32:37 And if there were any one place that I would probably want to put that, you know to me, we already talked about it a little bit. But the opportunity to create more flexible, more personalized hardship plans UH with deep awareness of whether we are increasing the loan value or not UH is UH is something that I just get continuously excited about. It. It's an immense challenge because the search space is so wide. You can change the length of the loan, you can change the interest rate of the loan, you can change the principle of the loan. You can do it when someone is in early or mid or late delinquency. You can offer it via email or text or call. There are so so many options in when and how to do this best, you know, but I think of this as really the most exciting thing on our horizon.

33:26 Yeah, I love I love that framing, like really thinking about the deep personalization of it, and I think, you know, certainly, I mean Jesse, thanks for joining us, giving so look and how Upstart is really turning Servicing into a true performance engine for the company. And I too, am excited about the idea of us being able to make this available to something that others can make use of as we continue to grow. Thanks Lene, Servicing. You know, it's not the thing that we dreamed of when we were little kids, but it's a pretty darn exciting place to be today. Thanks again to our listeners for tuning in to another episode of Leaders and Lending. Don't forget to subscribe and share the episode and we will see you next time.