New Article on Upstart

Good Article that identifies 3 main catalysts that could send Upstart shares even higher in the future.

  1. Increased Credibility will lead to lower rates for Upstart Customers. Banks are likely to identify Upstart as a reliable middleman in the future, which is likely to lead to a reduction in the cost of funding. As the model has been proven and is gaining more trust, partnering banks rates will continue to come down, leading to better rates and more loan originations for Upstart.

  2. The growth in popularity in Fintech will attract new customers and expand the valuation multiple as well as the general sentiment towards Fintech improves. Many Fintech companies have emerged as big winners in the last couple of years by solving day-to-day problems faced by consumers, and big banks have finally accepted the need for serious changes to compete with the growing popularity of Fintech. Median post-valuation of private Fintech companies have steadily increased in the recent past, suggesting that Fintech companies are growing fast and getting bigger with each year that passes by.

  3. Upstart is setting itself up to expand its first-mover advantages in certain demographics. As a Fintech company that caters to tech-savvy customers, Upstart’s strategic move to offer services in different languages might help the company carve out competitive advantages in the long run. The company now offers its products and services in Spanish, which was rolled out as a strategic move to serve the Spanish-speaking community in the United States that is often overlooked by banks and financial services companies.

Full Article:…


How is the AI used by Upstart different from Lemonade and why are they succeeding where Lemonade seems to be failing? Both seem to benefit both users and providers.


How is the AI used by Upstart different from Lemonade and why are they succeeding where Lemonade seems to be failing? Both seem to benefit both users and providers.

Well, to start with, offering actual insurance is different from powering others’ lending. Some of the basic data might be similar but personally I have to squint pretty hard to make Upstart and Lemonade look like they’re using AI in all that similar a way. If I think through the details in my “could I explain this to a curious elementary-school kid?” way, the first thing that comes to mind is that insurance risk (will Company have to pay for Loss) is somewhat different from credit risk (will Company not get back $ it loaned). They can be similar in that an unexpected event can result in a loss of company income and that they’d appear to serve younger folks, with Lemonade’s focus on renters’ insurance, but how each uses AI to create better risk models is going to be a point of divergence pretty quickly the more you dig. There’s also the interesting parallel that they’re using AI to disrupt VERY big industries with 150-year-old firms that could, at least hypothetically, throw a ton of cash at it and just do it themselves. And yet, as has been valuably pointed out here on this very board, those behemoths choose not to.

My first impression of Lemonade is that they were putting almost as much effort into their AI quotebot as they were into their risk modeling process.

Maybe the biggest difference is that Upstart is good at it, while Lemonade does not appear to be getting its footing easily. If I had to guess, though, I’d say insurance risk is shakier ground for an AI system to navigate than how likely someone is to pay back a personal loan. Or really a loan of any kind.

One Next Thing I wonder about, related to this, is how much of the AI structure/backbone these two companies will keep as they pivot to new products: Auto, Home, and Business in the case of each, and other kinds of loans (student, credit card) and insurance (who knows?). ISTR that the dialed-in risk parameters AI is so much better at estimating and incorporating will not be the same in those as for each company’s initial/tent-pole product lines.



How is the AI used by Upstart different from Lemonade and why are they succeeding where Lemonade seems to be failing? Both seem to benefit both users and providers.


I believe I can answer your question.

It is because insurance is a fundamentally different business of assessing risk.

Insurance: it is impossible to have risk-free insurance

Lending: it is theoretically possible to achieve risk-free lending


Risk is the only reason insurance exists.
For example: If there were a zero chance of you getting into a car accident, then why should anybody purchase car insurance? If risks of an accident were zero, the car insurance market ceases to exist!

But for lending, a borrower who has zero risk of default is both theoretically achievable and desired.
Lenders only want to lend to those who will always pay them back.
The borrower also wants a risk free loan because that means interest rates on the loan would be nearly ‘free’ to them- if default risk is zero, then the loan rate is merely equal to the lender’s desired rate of return (the time value of that money, whatever the opportunity cost is worth to be separated from their money for the duration of the loan).

A fantastic real life example of a “risk free” loan is very short term treasury bonds: The lender (which is you, loaning money to the gov) is charging the borrower (the federal gov) a risk free rate because you assume the gov will never ever default on its payment.

So how does this play into AI underwriting for loans vs AI underwriting for insurance?

This means that any algorithms or models you construct for assessing insurance risk can only ever be about correlations between variables, and not cause/effect. There is a fundamental limitation to the power of AI insurance underwriting; it can never be lower than a true statistical/probability rate of a loss event which must always be non-zero.

But for underwriting a loan to a borrower, it can be possible to construct models that describe equations of cause and effect which are more predictive of default than simple correlation models.

For example, if you know a borrower completed medical school then you know that can cause the borrower to get certain types of stable jobs/high income, and thus cause them to pay your loan back.

This simulated equation of cause and effect in that borrower, is analogous to a physics equation being near 100% accurate in predicting how long it takes a falling object to hit the ground, as opposed to a model that only finds correlations to generate a probability score (like X number of credit inquiries in the past 6 months is Y percent probability of loan default)

I refer you to UPST cofounder Paul Gu’s 2015 talk on this matter where he explains this better than anyone:

Timestamp 14:53 but I recommend sitting through the entire video!


the first thing that comes to mind is that insurance risk (will Company have to pay for Loss) is somewhat different from credit risk (will Company not get back $ it loaned)

Also, LMND underwrites their own policies. They do purchase reinsurance on a percentage to lower their overall risk, but ultimately they are responsible for all claims getting paid. In contrast, from UPST presentation, “97% of revenue – fees from banks or servicing with no credit exposure”. So it seems like UPST is a lot more insulated from the associated risks than LMND.


Articles like this trigger me. It is all noise. Even if the points are true and valid to some degree, they range from macro concerns to small details in my opinion (cater to the tech savvy?? More languages???). It is all very complex and intellectual, which are things I don’t value very much when investing (I’m talking about investing itself, not in areas like understanding the company product or business).

I’m (literally) willing to bet that all of these complex points are completely overshadowed by the simplest of points: Upstart is moving into more verticals, as fast as they can, starting with auto, which I read is something like 5x bigger and not contributing to earnings yet. Then there are mortgages, and I am sure more after that. That runway, or TAM expansion, combined with numbers that prove they are executing… well I don’t really need to know more than that. I don’t need to feel like the smartest investor in the room by spotting currents in trends or semi-hidden facts and angles. That is for the movies. I just need execution and growth. In fact, if that isn’t enough, it is a signal to sell.


Upstart is moving into more verticals, as fast as they can, starting with auto, which I read is something like 5x bigger and not contributing to earnings yet. Then there are mortgages, and I am sure more after that.


UPST currently has a job posting looking for a “General Manager of Product, Mortgage”:

Didn’t see this before, so I wonder when they will make an announcement that they are moving into mortgage?


It seems to me that the longer the loan period ,the more things can change. Thus it is less likely that the superior AI risk determination will work in long term loans like mortgages.

Another thought. Lending can be considered to have binary results, either the borrower pays off in full or he doesn’t. But nestled inside that is that some borrowers only pay off part of their loan , and a few default starting with first payment.
For instance if a borrower ha a long criminal record ,that increases the likelihood he will not make full payment because he will be in jail. FICO may give little insight into partial payment but obviously loan loss is reduced with each payment

FICO is 30 some odd years old but there are multiple tweaks out there for lenders to choose from. But AI brings a whole new dimension to risk determination. One need only look at 2008 2009 to see how bad bankers can be at judging risk using the old tools. Which probably are are not much better now.

1 Like

I am the first to say and understand that this promise of utopian lending is a fools dream. Lending is a cyclical market, and nothing is going to change that.

What UPST offers is a few points better results than a lender would otherwise be able to obtain both in defaults and in loans they can make. However, despite better results, in bad times all lenders are going down, UPST with that.

This said, who cares at present. We are in the midst of low hanging fruit and disruptive industry changing hyper-growth in the midst of a the beginning of a new economic up cycle (TBD given the inflation national debt, Covid, etc. but for the most part, ignoring these things have been the best way to go).

Thus, what I am saying is all the critique is largely true, but is not really relevant at present during this phase of growth and disruption. There is just too much good stuff going on. As we know things can change very fast, and if it does, like we saw with Alteryx, the numbers will speak volumes.

Alteryx is very instructive. The issues Alteryx had were recognized several years ago. (1) was its server client technology and no real cloud. Our retort was management said customers were not asking for cloud and the data was mostly local anyways. And given the low hanging fruit, it worked out. But something happened during Covid. (1) AYX underperformed the WFH stock appreciation materially. It ended up doubling, but only at its top, and otherwise was a strong underperformed in an enormous WFH market boom (during an economic crisis that otherwise caused most people to want to sell and panic (not us of course!)); (2) When people started working at home, it his companies in the face as to how backward server client actually was when the data was no longer local (As employees were not local).

So long and short there, had you thought that many years out you would have lost out on the incredible performance Alteryx gave us, plus, the numbers changed, we still could have got a double in 2020 on Alteryx if you were nimble, and the numbers told us when to get out and ruthlessly prune Alteryx from our portfolio.

At some point this will happen for Upstart (and Upstart may turn out to be more resilient than we think and a really long term winner) but for now, all we can see is what is in front of us. In the long run we all die, but that does not stop us from trying to at least feign to live a life in-between. It should also not keep us out of the stellar peak disruptive period of a hyper-growth dominant company just because it may be subject to future macro-economic conditions, and it may run out of low hanging fruit in a few years (it may not, but these things will, like with Alteryx, at some point probably have to be addressed and dealt with by Upstart).



Tinker one difference with UPST and bankers in the credit cycle is that Upstart is not a lender (direct loans are equal to only about $1 per share) The bad part of that is that they will not fit into the elite ,bailed out by Uncle Sam group if they get in trouble. The good part is they are not leveraged, at most all that will happen is that loam applications will dry up. Tough, but not enough to cause bankruptcy.
The financial industry is notorious for being a tech laggard and intuitively AI seems perfect for loans, so they may have an open road in front of them. The founders have been thinking about this and testing it for quite a few years ,from even before the software and hardware were up to heavy duty AI.
“it may run out of low hanging fruit in a few years” a few more years of similar out performance is enough for me.

My biggest worry is that they will be attacked for bias.

1 Like

However, despite better results, in bad times all lenders are going down, UPST with that.

Except that, other than a trivial percentage, Upstart is not the lender and so is not at credit risk. Moreover, if Upstart has done the good job of minimizing the risk of the loan pool, then the lenders that work with Upstart should be less impacted by an economic downturn than a lender who uses other methods of assessment.


Mauser, the bad part that Upstart won’t be bailed out in case of credit crunch is not a real bad point. Since the company doesn’t carry loans on its balance sheet it DOESN’T NEED TO BE BAILED OUT as losses on existing loans would’t impact its balance sheet.

I agree with Tinker that in case of downturn in economy and subsequent credit crunch Upstart would underwrite less loans hence will be getting less fees as revenue. In this respect we have a cyclical business which is not equal to our typical SaaS model. Again though in agreement with Tinker I think that we are in the beginning of the up cycle in the economy and there is a relatively high probability that Upstart could end up making close to 1b in revenues this year as mentioned by Chris/GR and possibly close to 2b next year. In such case today’s share price of around 200 will look like a gift next year. This is speculation though, let’s stick to the numbers.

I’m actually these days reading a book about Goldman Sachs and I see a parallel between typical investment banking type of underwriting larger loans (fixed income) and selling to institutional and professional investors and what Upstart does for retail personal loans. I think that it’s a very good business model from risk perspective - u take fee and u don’t take the loan on ur balance sheet. Compare it to, for example, typical fintech/neobanks like Sofi which takes personal loans on it’s balance sheet. Secondly, this underwriting business is strongly profitable as evidenced by results of investment banking divisions of Wall Street firms and own Upstart results - when I looked at Upstart for the first time in Feb/March this year I was positively surprised by the bottom line numbers and each quarter is getting better and better. Third observation - heavy lifting in Upstart business model is done by tech and AI as opposed to star bankers and rainmakers with investment banking departments. So, scalability of the business is enormous compared to scarce resources of IB rainmakers on Wall Street.

I’m not sure how helpful is this comparison of Upstart’s business model to investment banking’s underwriting business but the fact is that Upstart got a very smart and profitable business model as it outsources loan related risks to third parties.

Long UPST with around 19% position
Sold SOFI after disappointing ER this week


Clearly the credit business is cyclical, and a down cycle in either demand for loans or willingness of lenders to loan will not be good for Upstart. But not catastrophic either.

We seem to be in a minor cycle of reduced loan demand at the moment. Not for the usual reasons of a crash but because of near helicopter money being pushed out to consumers. They are flush with cash and covid restrictions have reduced some of the ways to spend it. Ergo less desire to borrow.
Will the super generosity of the Fed and federal government continue at this rate forever? Probably not ,but I never underestimate vote buying .Even then it is more likely than not that credit demand will pick up over the next year.

One would think low interest rates would stimulate loans . But most consumers do not get low rates, credit card rates are outlandishly high.

Upstart TAM is huge. I am particularly optimistic about Upstart giving loans to Hispanics . Who are mostly ignored by main stream institutions. In the past Hispanics tended to be more debt adverse than Whites, so it is a wide open market chance


The fact that UPST is doing so well despite a period of reduced loan demand is a positive imo. Also, consumer credit demand is picking due to several factors including ending of the enhanced Unemployment benefits, reopening, and even price inflation of everyday items. I can see UPST originating more loans and as more lenders get comfortable with the platform, I see more partner banks providing funding especially with such low rates on home loans and CRE. if the loss rates are comparable, load up the folio with consumer credit.

1 Like