Upstart CEO Dave Girouard: triple-digit growth, but US consumer stress is at multi-year highs

joining the show. Thanks for joining. We'll talk to you soon. Cheers. Bye. Up next, we have Dave from Upstart. Do you know what Do you know what Upstart is, John? What What is Upstart? I can't hear you. What? What? Oh, sorry. There there seems to be an air. Well, we'll hear it from Dave directly. Welcome to the stream.

How are you doing? Hey, good to be here, guys. Sorry to keep you waiting. It's great to see you. What does Upstart do? Everyone's been asking, people want to know uh what do we do? It's a great question. We uh we are a lending platform. Yes.

So, we apply AI and machine learning to consumer lending and uh we we operate in the form of a marketplace where we have consumers that we market to on one side and all sorts of uh banks and credit unions and private credit and all sorts of sources of capital on the other.

And the whole uh basic premise of the business is to apply AI to the foundational notion of of making consumer credit work both in terms of origination and servicing etc.

So yeah what what in the I mean when when most people today say AI they mean large language models they mean post chat GBT but obviously you've been in the business for a long time and machine learning is been a relevant technology pre-transformer-based large language model.

So how is AI in the modern context of like the large language model, the generative AI context, how is that changing your business and uh or or is it more of like a sustaining technology from you for you as opposed to like upending everything that you do?

Uh well, you know, I think uh a AI in in many forms in in LLM's in in sort of that sort of generational notion of AI obviously has grabbed a lot of attention, but when you think about, you know, high frequency trading, genomics, uh medical imaging, autonomous driving, these are all like forms of AI that that you would not they're not language based, they're not LLMs, but they are of course changing things pretty rapidly.

So I think you know maybe the big question is is there a unifying future where all this comes together into some form of you know AGI but regardless of whether that is true or not we are building something that's different it's foundational in nature meaning all the data on our platform is created by our platform which is you know very different than how LLM's work um but I I would say the commonality is that look there's just enormous win that machine learning and AI can bring to any particular task at hand in our It's making a consumer loan of of many forms much much better.

And that means like zero process, perfect pricing, works for the lender, works for the borrower. And uh you know, we we started we were founded, you know, 13 years ago.

We didn't really use the term machine learning or AI until 2017 when we kind of felt like this what we were building was sophisticated enough to warrant that name.

But, you know, it was all under the covers, you know, no one thought much about it until uh, you know, chat GPT and and November 22, I guess it was when the hot overnight success 13 years in business. Loved it.

Yeah, I I imagine as you've seen the advances over the last couple years, every time there's a new model release or or even even a vendor that's saying we're going to help you like better process PDFs, I imagine that that's exciting to you because you guys have done the heavy lifting to like build the supply and demand.

And so as new technology emerges, you can just help uh you know make you know make that process more more and more efficient.

How how much like what's your decision-making process around um you know whether you you guys want to build something in house which I'm sure you were forced to do a lot more maybe pre2020 to to now when there's a bunch of um new infrastructure providers that that you guys can leverage. Yeah, it's a great question.

I mean we've always built everything inhouse. Uh when something looks pretty obviously commodity like and that it's on top of LLM.

So for example what exactly what you mentioned extracting information from a document not not just kind of OCR but actually understanding the context that information in in a way that you can take all this human effort out.

Now that's something that honestly is very commodity like meaning the prices we'd pay aren't much different than we would pay if we built it on top of one of the LLMs. So we're always like looking for things where we if we can ride someone else's cost curve on on some commodity that's great.

We are definitely trying to build the larger picture. You know, the the the sort of endgame for us is if you can imagine 100% of Americans are permanently underwritten. They can have any form of credit at the very best and guaranteed best possible rate in a moment with no process whatsoever.

So, anything that sort of gets us closer to that quickly. And uh there's definitely you know business models evolving on top of LLM that I don't you know it's not my problem to figure out whether they're sustainable over time.

All I know is is like okay if you want to charge me an extra 15 cents you know that's great and take care of all these logistical problems of maintaining that particular specific small model like the the kind that you referenced. Uh how are you thinking about the top offunnel evolution?

We were talking earlier in the show about um Google versus Chat GPT uh the GPT5 launch and the model router and it feels like in the future you might be able to go to chat GPT and say I need a loan and you are probably going to be there. Are you thinking MCP servers? Are you thinking about SEO in LLM foundation models?

Like how are you thinking about the changing landscape on the top of funnel? Yeah, you know it's a great question.

I I was eight years at Google before I founded the company and There's a lot of history and there look at some Google or maybe one of the others are going to come to us and they're going to say we want our agents to be able to apply for loans and we don't want you blocking it or whatever. How do you feel about that?

And I've said to our team, you know, I'd rather we do that before they do that. So, you know, maybe the question is as these agents evolve as as as true agents for the consumer, you know, I'm super curious. I don't necessarily know the answer.

Will there be three or four of them from the from the giants out there or will there be plugins to those to handle much more domain specific tasks or things? I don't know how that will evolve, but I do believe it's without question you're going to have somebody that will do that on your behalf.

It will get the best possible outcome for you. Hopefully, it will also help you make better decisions. All right? Not not just go through the the mechanics of applying for a loan, but help you really understand like what's the best product for you? Should I even be taking out a loan?

If so, what what other choices could I make? So, that kind of stuff we we are working on for sure. We could just think of as the sort of agentic part of this. Uh and we would rather be uh we'd rather be applying that to others than having it apply to us. For sure. Of course. Uh last question for me.

Um uh obviously you see a lot of consumer economic data. How are you feeling about the health of the American consumer right now? Yeah, I mean we watch this a lot. We we've built an index to sort of track it.

We call the upstart macro index, which is really about like the health of the American consumer and how that's impacting credit performance.

So basically what you see across all forms of credit from cards, student loans, you know, auto loans, mortgages is highest default rates that they've seen in a very long time since pre prior to COVID.

So the consumer has been stressed and is stressed and maybe it's inflation just overspending you know habits built during co that uh for spending that haven't dissipated you know so so the consumer is definitely stressed it's been priced into our model for a very long time so we're very calibrated to it but I think you know we you have begun to hear if you you know retailers are seeing people pull back they're being more choiceful about what they're spending money on suddenly just all across the board you're getting a lot of a lot of noise out there, same store sales being down for different types of industries.

So, I think the US consumer is finally kind of going, "Holy [ __ ] you know, we're not earning as much as we thought we are and we're spending more and, you know, this kind of we have to get back to a normal place.

" The the thing we I point at more than anything else is the personal savings rate, which is something, you know, produced by the government and that's at almost historic lows. So, you know, people are not saving, they are spending and and and it's, you know, a bit of a a catch up that's needed.

So from our point of view like a little bite of recession if it comes down to like consumers slowing down and you know spending less of what they earn like would be a good thing from our perspective. Makes sense. Yeah. Any any any comments on uh I mean obviously everyone's been debating you know stock markets ripping.

So if you're just looking at that doesn't feel like there's a the a real reason uh to lower rates but if you look at some of the you know employment data and the data that you're talking about like maybe there is real argument to lower rates like what's your guys's like internal outlook uh for for the the back half of the year and beyond.

Yeah we you know in terms of our real product and what it's projecting we we never project changes if you will so it's always based on what the rates are today.

Having said that, I mean, I we are certainly I I I think rates are unnaturally high considering where uh inflation is, which is really kind of, you know, the things they have to weigh against. So, in my mind, they're they're likely to move down.

Uh I I I can't predict the impact of of all these tariff stuff on inflation any better than anybody else, but I think generally speaking, you know, the the rates should probably be 100 or 200 basis points lower. I think they inevitably will be lower.

They're not going to go back to what they were, you know, in 2020, but but they're going to go a lot lower. And that's a tailwind to our business.

We again, we don't plan on it, but there's definitely a point at which the consumers are going to get in a better health position, saving more money, rates are going to come down, and and all that is, you know, future tailwind for us. Yeah.

Feels like you're really set up well for the next couple of years, like built through, made it through high interest rate environment. If interest rates come down, you're you're you're ready to rock. So, congrats on all the progress.

No, it's I mean, we just reported tripledigit growth in our earnings, you know, last week. The market hammered us anyway. Well, you're 5 years in. You're you're almost a veteran now. 5 years. I feel like if anyone can take a hammering, it's you. This is foric growth. Let's hit the gong. Thank you. Congratulations.

Thank you for coming on the show. I'd This is a great conversation. I'd love to talk to you again. Thanks, J. Have a good one. Cheers, Dave. Let me tell you about eight sleep. Get a Pod Five. Fiveyear warranty, 30 night risk-f free trial, free returns, free shipping. Jordy, I think I beat you. What's your number?

And my problem is I I get like four great night sleeps in a I got an 81. How'd you do? 94. Play the Ashton Hall sound for me. Let's go. And we got a question from Bill Bishop, uh, who I'm a huge fan of. He writes, uh, Cynicism on the Substack live stream. He asked, "Shrooms and chat GPT, good or bad?

" I say, "Absolutely bad. Stick to the classics. Caffeine, baby. That's all you need. What do you need shrooms for? Cheers. Stick to Quick. Cheers for Bill. Stick to diet. Matina from Andrew Huberman. Load up on the caffeine. They're calling it a podcast. Have delusions of grandeur.