Featured Guest: Jeff Keltner

Jeff leads Upstart’s efforts in strategic partnerships and new initiatives – and is the host of their weekly podcast Leaders in Lending. Jeff joined Upstart in 2012 after spending 6 years at Google. He launched and built the Google Apps for Education business, growing market share from zero to almost 70% in its first four years. Jeff spearheaded marketing efforts for Google Apps in Global 2000 accounts and led sales, business development, and go-to-market strategy for the launch of Chrome devices in the education and enterprise sectors. He spent several years in direct sales at IBM, always exceeding quota, and was a founding engineer and lead UI developer at SSB Technologies. Jeff holds a BS in Computer Systems Engineering from Stanford University.

Episode Transcript

0:00 – [MUSIC PLAYING]

0:03 – Jeff Keltner

I think when we say, AI everybody has this kind of like in ChatGPT. It plays into this like a human-like interface where it feels like I’m interacting how The Terminator or whatever, your favorite, The Matrix. But it feels like I’m interacting with some sort of a human-like intelligence. I think when people hear AI, that’s what they think. A lot of AI today is really very sophisticated prediction algorithms but they’re just prediction algorithms.

0:28 – David Reiling

Welcome to the NextGen Banker podcast where we explore what’s next in banking and talk with the innovators responsible for creating positive change in the financial sector, I am your host, David Reiling and I’m very excited to welcome Jeff Keltner today. Jeff, thanks for being on the NextGen Banker podcast.

0:43 – Jeff Keltner

Thanks for having me. I’m looking forward to it, Dave.

0:45 – David Reiling

Well, Jeff, right just before we get started, we always throw out a little reminder to our audience to listen to the musical feature at the end of each episode. It is somewhat of an interesting– we got a little cult following with us on the music. So, every episode, we showcase one new artist from somewhere around the world and representing a wide range of genres. So, check it out. Sometimes pretty cool. We got a pretty good following there.

1:11 – David Reiling

So, Jeff, a little bit about your background, you are the senior vice president of business development at Upstart. And before joining Upstart, you spent six years at Google, several years at IBM working in sales. You certainly are no stranger to the podcast world as you host leaders and lending where you talk about the lending world in the future of the lending industry. Well, I’m always curious with my guests how they get where from their start to where they are today. And if I have it, you started out with a degree in engineering, then how did you end up in the finance industry and Fintech?

1:51 – Jeff Keltner

We call the word, I heard something of an accidental banker, accidental Fintech. So yeah, I got a degree in computer systems engineering, which is some strange mix of electrical engineering and computer science to 20 kids at my school do every year, and somehow I got roped into doing that. I don’t really know how that happened. And when I came out, I ended up deciding that I wanted to spend my career at the intersection of technology and what it can do in other industries, where it makes an impact. So, I spent a couple of years at IBM selling, believe it or not, mainframe type computers. It has for hundreds of times is an old school tech. And then I joined Google right in time for the launch of what is now Google Cloud. But Google, it was originally Gmail for domains, Gmail for businesses. And I ran the education side of that for a number of years and then ran a lot of the global account strategy as we were trying to figure out how to displace Microsoft in those accounts.

2:41 – Jeff Keltner

So really again, at that kind of– for the cloud revolution, how does a cloud get deployed in different industries. And then the individual that had been running the Google Enterprise division, Dave Girard, left and founded Upstart. And I called him and said, this place feels pretty big, and stodgy, and slow now. What are you doing over there? It sounds interesting. And so, he was really looking at the problem of access to credit and how modern technology could expand access to reduce the price of credit in a variety of ways. And so, I jumped on as the fourth person in the door, I think. And it’s been a heck of a ride ever since.

3:10 – David Reiling

Yeah, that’s fantastic. It sounds like it’s a heck of a ride. So, let’s jump in to where you are today. So Upstart is an artificial intelligence or AI marketplace, straight off the website is to improve access to credit while reducing risk and cost for bank partners. So maybe could you just kind of break that down a little bit for us? What does that mean in real people terms?

3:34 – Jeff Keltner

Yeah. So, the core technology that we’ve developed are really a couple. One, kind of the onboarding experience, initially just for personal unsecured loans. So, think like $10,000 or $20,000 term loans, usually three-year or five-year length loans. So, it’s kind of a nontraditional bank product. And banks have had it, but it’s never been a very large portion of anybody’s asset base. And I said, how do we do digital onboarding? How do we do sophisticated underwriting for that? And that’s kind of the core technology we developed. And the way we go to market is really a twofold strategy. One, we license that technology to banks and credit unions who can use it within their own customer base, who can market the loans and just use this as a kind of a loan origination experience and an underwriting capability to help them approve more customers for this category of credit.

4:20 – Jeff Keltner

But the second part of that is also, many of them did not know how to drive demand for this new product. And so, the marketplace component is where we actually have a consumer facing site. And consumers come and they can submit their information. And we’ll check it against our lending partners and match them with one of our lending partners that would have offers available for them. So, they’ll end up originating a loan with one of our bank or credit union partners, even though they found the opportunity through the Upstart site. We found that was really useful for our bank and credit Union partners because many of them wanted more loans. This is a few years ago. Now, they’re not all asking for more loans in the current macro environment. And they weren’t always effective at driving them. So, they say, can you help us– we love your loans, can help us get more of them than what we’re generating on our own? So, we ended up kind of building both the ability to power their experiences as well as to drive consumer demand for at least the products we have. And we’ve now expanded into auto purchase, auto refinance, and we’re working on other categories of loans to apply the same technology to.

5:13 – David Reiling

That’s fantastic. And so just from, I’ll say, for the financial institution side, does a bank or a credit union just license the technology and you have the underwriting box there? Or is there any input with that institution?

5:26 – Jeff Keltner

No, that’s a great question. The underwriting is always, I would describe it as being owned by the bank or credit union. And we have a sophisticated risk model. But that risk model is making a prediction, likelihood of repayment, likelihood of default, timing of default. How that prediction turns into an offer of credit in a price of credit is entirely at the discretion of our lending partners. Are they willing to lend to– they might have, in addition to our model, they all have, we don’t have go below a certain credit score. We don’t go above a certain debt to income ratio. We might not go below a certain income threshold. We might not take people who are in bankruptcy or had recent delinquencies, things like the very traditional credit metrics.

6:02 – Jeff Keltner

This is our box. They say, great, we will make sure that our system will respect that box and make offers of credit respective to your credit policy. And then the pricing strategy, if there’s a certain level of risk that Jeff Wheeler won’t repay a loan, what kind of offer do you want to make him? And that’s totally at the discretion of each of the individual lending partners to the same consumer with the same Upstart risk models underneath. The same way you have the same credit score through a lot of bureaus, but yet every bank has a slightly different way of looking at how they want to price that given risk. And so really, there’s a lot of control we put in the hands of our bank and credit union partners to make sure the program matches their objectives, their goals. Maybe they want higher yield, more prime customer. Maybe they want to reach into a more risky segment where they can make a higher yield. Maybe they want to just bring customers on at a low price. Whatever their goal is, the system is configured to do that for them. And the sophistication we have is just in helping them understand the risk that they might be taking on when making a given offer to a consumer.

6:56 – David Reiling

Gotcha. So that’s very intriguing. And so, if I had to pull you then into maybe the magic of what is Upstart, let’s talk about the AI. Now, I am kind of an AI nerd. I have I have no formal background other than at least once a day, if not more, I’m on ChatGPT or OpenAI. And I’m just testing and trying it because it’s fascinating to me. And that’s in a lot of ways people’s view of artificial intelligence at the moment. But specifically in what Upstart is doing in making impact in the finance industry– gosh, I guess there’s two camps here. There’s those that are very skeptical about artificial intelligence. It’s a scary thing. But you’re using it in a way that seems to be very beneficial. So how are you using it across the banking industry?

7:48 – Jeff Keltner

Yeah. I’ll give you the two places we use it. But I do think it’s important for the audience in this segment. I think when we say AI, everybody has this kind of like in ChatGPT. It plays into this like a human-like interface where it feels like I’m interacting how The Terminator or whatever your favorite, The Matrix. But it feels like I’m interacting with some sort of a human-like intelligence. I think when people hear AI, that’s what they think. A lot of AI today is really very sophisticated prediction algorithms but they’re just prediction algorithms. They’re far more complicated than a linear regression. They can look at far more variables. They can look at how variables that are highly correlated might still be additive when you look at all of them together and understand how those correlations play, and what it means to have a high credit score and a high debt to income ratio. How do I weigh those off in the context of a given borrower? Those kind of questions, AI is really good at answering. It doesn’t feel anything like talking to ChatGPT, which is a fascinating experience.

8:40 – Jeff Keltner

And so, we’ve really looked more at this kind of back-office side of AI. And I think if you look at the banking industry, you’ll find that AI is not– it’s here, it’s happening. It’s just a question of where is it being applied and how quickly is it being applied to other segments of what gets done. When will there be a virtual teller that’s ChatGPT powered is a fair question. But if you look at what we do, we really focus on three areas, I’ll say right now, maybe four of where we can apply AI in a more back-office kind of way that just enhances capability. So, the first is understanding the level of risk. We can take over a thousand variables from multiple credit bureaus, extra variables, information about an applicant being provided by the applicant, their level of education, their occupation, their field of work, that kind of thing. And we can make a much more accurate prediction of risk.

9:27 – Jeff Keltner

And the reality is that for most Americans, their credit score does not really represent their level of risk, certainly not across this. I mean, you get the same credit score for a $1,000 loan and a $1,000,000 mortgage. That doesn’t really make sense. And it turns out that more than 80% of Americans have never defaulted on a credit obligation and yet less than half have a prime credit score. And so, we’re trying to say, how can we use larger sets of data, more sophisticated algorithms to identify that other large portion of Americans that are going to pay back a loan and help you understand that risk. So that’s the first area we apply. And that was kind of the founding belief. Many more Americans are creditworthy than we think. Can we find them with machine learning? Can we look at as opposed to 20 variables in a traditional credit bureau– or sorry, bank or credit union might look at? Can we look at 1,000 or 2,000, and by putting those all together with an AI algorithm, understand the risk level of a borrower much, much better? The answer to that has been a resounding yes.

10:16 – Jeff Keltner

The second thing that we came upon, I won’t say by accident but through the journey, was the importance of simplifying the onboarding experience and removing friction. So, when we started the process with our bank partners, every borrower went through a phone call, some sort of knowledge-based authentication, some sort of ID ver– some sort of document upload. Give me your driver’s license the way I might have done in the branch, just going to upload the photo. And then we had this question that occurred one day and said, well, what if we could not ask for anything? Like what would happen? And so, we took 2% of borrowers with less than $5,000. You know, this is really small subset and we said, don’t ask where we feel really good that it’s not fraud, where all the signals seem positive. What would happen? And what we found was that when you went from upload a document or talk to a person to a fully automated experience that could be done in one sitting, maybe it’s 20 or 30 minutes but you’re not leaving the browser while you do it. You start it, you finish it, the money’s on the way when you’re done. We could between double and triple conversions.

11:09 – Jeff Keltner

And then we thought, wow, if you can double or triple conversions, how do we go from a small portion to a much larger portion of the applicants to whom we can provide that experience? And so we started applying AI to go, how can we figure out the smallest number of things we can ask a customer to do to get confident that all the things they told us are true, that they are who they say they are, they do make them out of money they said they make, that they did have the job that they said they have? And when can we reach the threshold saying we’re now confident in issuing the loan to this person given those factors, without maybe asking for a driver’s license upload? Now, sometimes you’ll have to. And sometimes we ask for a video selfie of a borrower with their driver’s license telling us how much they want to borrow and what they’re doing with it. Because fraudsters don’t want to do that. And if we have a lot of concern about fraud, that might be helpful. But you don’t want to make everybody does that. And so, we started applying AI.

11:57 – Jeff Keltner

Now more than 75% of the originations through the platform have zero touch by a human through, not just the underwriting but the full verification process all the way through origination. And that really increases the throughput. So, if you think about what we started and how do we assess risk, this is kind of the same thing. How do we assess risk of misrepresentation? The other areas we really apply AI now are how do we best target marketing outreach so that we get the highest propensity customer who’s going to convert at the lowest cost. There’s a ton of efficiencies to be gained there. And then probably the most nascent area for us is, where do you see the most risk in your current portfolio? How can you best reach a customer who might be at risk of going delinquent, who maybe just went delinquent? How can you best identify the need for a hardship, maybe even get ahead of somebody with a hardship offer when you know that they might be at risk? Those kinds of things that allow you to improve your servicing outcome. So, kind of across a lifecycle, we see lots of ways to apply these sophisticated algorithms to improve your understanding of what’s happening for your customers and to improve your ability to serve them. And so that’s where we see a lot of that going on. And none of that is like the chatbot. None of that is like replacing a teller. It’s all kind of back-office type stuff. But it’s amazingly powerful in how it can change the economics of the lending business.

13:03 – David Reiling

Yeah, definitely. Especially, I mean, that user experience going into it is that first impression in terms of the loan process. And somebody who’s going to go into that loan process, they’re looking for a yes where they may not get it. But the fact is that the experience is good enough. It doesn’t completely deter them from coming back in the future.

13:23 – Jeff Keltner

Well, I’ll tell you the other interesting thing. You talked about that deterrence. And one of the questions we had when we saw this high uptech was like, well, when we stop asking people for verification, are we going to decrease the quality of credit because people are going to take advantage of us? And our initial results were the exact opposite. We improve the quality credit. We went, OK, why? And what we found is that the good borrowers are not only rate-sensitive, the good borrowers are also more effort-sensitive. And they go, I’m a good customer, why are you making me upload 15 docs? I’m going to go somewhere that’s going to treat me better, not just from a rate but from a process point of view. And so, these things really are important. If you want to have a good relationship with that customer, you’ve got to give them a high-quality experience, which means a simple, easy experience to the greatest extent that you can. You can’t do that for everybody. Now, there’s risk in the game. And that means we have to be careful and put friction in the process where we see high degrees of risk. We don’t want people– we don’t want fraudsters coming through the system and taking advantage and hurting people’s credit scores. But it is a really important thing, not just for the business but for the customer. Because that’s what they’re looking for.

14:23 – David Reiling

Cool. So, Jeff, from the standpoint– I’m going to narrow you down a little bit in regard to. So, Sunrise is very focused in regard to serving people and populations that are generally underserved. And a lot of times, they’re low and moderate income or they’re immigrants. And they’re generally those that are the least likely to enter the system in the traditional banking world, if you will. How do you see AI being used, maybe in those circumstances or for good, to get that person their first loan or build that credit history for the first time?

14:55 – Jeff Keltner

I love this question because I feel like there’s so often this question about bias in AI and is AI going to discriminate and take systemic bias that exists and just exacerbate it. And there’s always that risk. I don’t want to say it’s not possible. But I think it understates the problem that faces these communities today. We’re worried about preventing that but we’re also blocking off the ability to do exactly what you’re asking about. Can we actually make the world better with these technologies? And I don’t want to stand in the way of that too much because I think there’s a ton of opportunity. So let me give you, I think, the two areas where I think this really can make a difference. A, I kind of gave you that statistic that many more Americans are creditworthy than have prime credit scores that would get them traditional access to a bank or credit union type credit. And they end up turning to all sorts of alternatives, buy here, pay here car loans, or payday loans or title, all sorts of stuff that’s certainly not as friendly to the consumer.

15:45 – Jeff Keltner

And I think AI can really help us find those customers who are credit worthy. There certainly are fewer of them, percentagewise, in a lower income or a traditionally disadvantaged community than there might be in a more prime community. But at the same time, there are many of them are there. The majority of them are good customers. And if we can help you find those and sort the good from the risk, then that helps you serve a broader population. Because what does every bank and credit union do when they can’t identify the risk well? They just cut it off. They go, oh, we’re just going to have a 700 minimum credit score. And then I’d much rather say, let’s go down to 600, or 580, or 550. We have some of our partners that have no minimum credit score, and they rely on the risk algorithm to find out. Does that mean they serve as many people below 640 as above? Of course not. There’s more risk there and they can’t approve as many, but they can approve some. And they can improve a lot more than they could in a traditional world. So, I think there’s a ton of opportunity to expand the approvability without increasing risk for the institutions just by understanding which members of the lower credit score communities are actually worthy of credit, are actually good risks to take. And there’s a lot of that mispricing, misrepresentation by credit scores in the markets, tremendous opportunity.

16:50 – Jeff Keltner

The second area, kind of related to that automation I talked about, is a lot of products that these communities need are not the bread and butter of traditional banking. They’re not new car loans through franchised dealers. They’re not mortgages. And often the cost of offering an alternative to a payday loan, a $1,000 loan or a $500 loan– why does the bank or credit union not offer a $500 loan? We’ll go, well, Jeff, by the time they walk in the branch, talk to my guy for 10 minutes, and the underwriter looks at it for 20 minutes, I’ve lost money, even if none of them ever default. I’ve already lost money; I can’t do it profitably. But that’s because we have a very person-centric view of what that process is. And if I can move to that automated underwriting, automated verification, I’m really reducing the cost to originate. And it opens up the ability to serve, not just the personal unsecured loans which as we said, even a $10,000 loan was something very few banks did before, but a $1,000 loan, a $500 loan. That is really the kind of product that many communities need access to and don’t get it from the traditional financial system.

17:52 – Jeff Keltner

And if we can leverage that AI to lower the cost, the barrier to entry, I think we’ll see many more banks and credit unions that want to serve that customer, they want to provide that product, they want to help the community. But they just can’t do it without losing money and they’ve got to find a way to do that. And I think AI can solve both how can I find the right people to lend to and how can I reduce the cost to originate where I can put these kinds of products in market, maybe not with a ton of profit, but not losing money. And most of the banks I talked to would be happy to offer a payday alternative if they just didn’t lose money. They don’t need to make as much money as they do on a mortgage. Because if I can just deploy my assets to serve my community and break even, that would be good for me. And I think we can make that happen with AI. And that’s, to me, a tremendous opportunity.

18:29 – David Reiling

Yeah, that’s fantastic. And that’s where we are so on the same page. I preach this regularly, that how can we really use this technology for good and open up access to markets. And ultimately, that access to capital, even if it’s small or micro, drives a community at its very base level and it’s good for everyone. Because at the end of the day, there’s going to be a merchant on the other end who’s going to sell something and can make a profit, and is going to buy, and going to grow, and so forth. So, it really is a critical capital for those underserved communities.

19:00 – Jeff Keltner

Yeah. And everything we’ve been saying applies equally from– I focused on the consumer side, it’s where we’ve been, but small business lending, which is the bread and butter of how you start businesses in a community, and by the way, one of the hardest kinds of lending to underwrite and operationalize at reasonable cost. And that’s why so much commercial lending is large commercial real estate because it’s a large transaction. I know how to write it, there’s a secured asset on the other side. I know how to do that. If you get down to the guy who needs $50,000 to open a restaurant, I mean, it’s hard just to verify that he’s got a lease on a place and is going to open the restaurant. And so just how you get to understanding the risk of that better so that you can approve more people and how you understand how to operationalize that without sending somebody out to the address to look– that’s a traditional way of it. Is there really a business there? Is it operable? And somebody have to look. Well, if you can get past that, then to your point, the lifeblood of the community is small businesses and the lifeblood of any business is access to credit. And the more we can expand access to credit for consumers for small businesses, the more we can really serve our communities.

20:00 – David Reiling

Fantastic. So, Jeff, I have one last question for you. Now, you mentioned the AI chatbot teller and so forth. But the basis of the question is one that we ask every guest, and that is what do you think the next generation of banker looks like?

20:15 – Jeff Keltner

ChatGPT in a tie, is that supposed to be the answer?

20:17 – David Reiling

That could be it. That’s I’m headed, I think.

20:20 – Jeff Keltner

I mean, I think the nature– I don’t know, I’ve never been a banker per se. I’m an accidental banker. So maybe I’ll just say, I think the skills that are going to be critical for the banker in the future are going to be a high fluency with data. People sometimes describe data as the new oil, and I think it’s a terrible analogy. Because the oil is valuable because it’s rare, it’s hard to find, and only a few people have it. Data is going to be the opposite. Everybody is going to have it, but it’s going to take a lot of understanding and effort to get out insights from it. So, I compare it more like the new sand, it’s everywhere. But turning the sand into the silicon is the magic. Everybody’s got the sand on beach everywhere, but it’s that process of refining the sand into something more valuable, of turning it into a useful asset for the institution, that’s really important. So, I think every banker at every level is going to have to become with a high degree of data fluency and understanding of that kind of stuff.

21:09 – Jeff Keltner

And then this sense of technology, technology is going to intermediate a lot of interactions. And having an understanding of how those technologies work, how AI works– not that you have the right models, not everybody’s going to be a machine-learning engineer, but that you can look at the business and go where can we apply these capabilities to provide a great experience. And how do I best meld that experience with the in-person experience or on-the-phone experience? That customers still want; they still want that. And so, I think the banker they can understand where the technology can do its magic, how to integrate it into an also human interaction experience and can make sense of the volume of data that will be available, those will be the bankers that are able to provide the most value to their institutions. And so that’s what they look like. But that’s the skills I think that the successful bankers are going to have moving forward, is understanding of that dual modality of the customer, understanding of the capabilities of technology and how to apply them best to their business in a real fluency with data to make those kinds of decisions, not based on just intuition but based on the information you have and how you can make best use of it.

22:10 – David Reiling

That is fantastic. So, Jeff, it has been a pleasure to talk to you today. And we appreciate your insights, particularly, because I share that same passion of that intersection of, I’ll say it’s commonly the high tech and the high touch type of aspect of where we see banking going and the particular, whatever it means to be a banker in the future. There’ll be some blend of that together. So, thanks again. We appreciate your time. Thanks for being with us. Thanks for listening to the NextGen Banker podcast and we’ll see you soon.

22:41 – Becca Hoeft

For this episode’s musical feature, we’re showcasing Adrian Walther. Most of Walther’s music consists of minimal piano and acoustic instrumentation. You’ll always find an emotional theme throughout his catalog. Here is “Baby Knows” by Adrian Walther.

22:57 – [MUSIC PLAYING]

23:39 – Becca Hoeft

That was “Baby Knows” by Adrian Walther. You can find more of Adrian Walther’s music on Spotify. If you would like your music featured on the NextGen Banker podcast, email david@nextgen-banker.com with a link to your music and website. Thanks for listening to the NextGen Banker podcast. We’ll see you soon.