Upstart and customer concentration risks

I am a silent lurker and a learner and have been following this board for more than a year. I should say that I have a learned a lot from this wonderful board and thanks to all the contributors.

I have a close to 10% position in Upstart and so I will be paying attention to what they have to say when they report on May 11th. Upstart (UPST) is cloud-based AI lending platform. GauchoRico introduced this company and recently Saul in his monthly update took a position in this company. This company is founder-led, disrupting traditional lending business. Their founders hold approximately 20% of the company. Their AI powered lending platform incorporates more than 1000 variables to automate the lending process unlike using a single metric like FICO scores. Their platform connects consumers, banks and institutional investors. Customers like their platform as they have 4.9/5 rating on Trustpilot. Although they acquired Prodigy software earlier this year, a first mover into automated process of originating and underwriting auto loans, Dave Girourd, CEO of Upstart stated that any revenue from this acquisition will probably not contribute to 2021 revenues.

See recent Upstart’s deep dive by Brian Feroldi and TMFstoffel.

See Saul’s April, 2021 portfolio update…

This is directly taken from Saul’s April end Monthly Report-

“Quarterly revenue really took off in 2019, and the last eight quarters look like this. (You can spot that Covid quarter a mile away):

20 33 49 63
64 17 65 87

And they are guiding to $115 million this quarter, up 80% yoy and up 32% sequentially!!!”

They recently signed Drummond Community Bank for personal and Auto loans on April,28th… and

First National Bank for personal loans on April, 14th.…

Customer Concentration Risks-

I think one has to keep in mind that the 2 biggest risks for this company comes from Cross River Bank (CRB) and Credit Karma. CRB accounts for a significant portion of their revenue. From their 10-K “Cross River Bank, or CRB, a New Jersey-chartered community bank, originates a substantial majority of the loans on our platform.”

“CRB, originates a large fraction of the whole loans sold to institutional investors under our loan funding programs. For the years ended December31, 2018, 2019 and 2020, fees received from CRB accounted for 81%, 80% and 63%, of our total revenue, respectively. We have entered into a loan program agreement that governs the terms and conditions between us and CRB with respect to loans facilitated through our platform and issued by CRB.”

Although it is encouraging to see that revenue contribution from CRB dropped from 80% to 63% from 2019 to 2020, it still is a potential risk. CRB on the other hand from what I learned is unlike any other community banks and has partnerships with other lending partners like Affirm and Rocket loans. But again, Upstart is just starting out to expand its business and probably it is normal to have significant customer concentration in the initial growth phase.

Another potential risk that was also highlighted by Bert in his article is the traffic from Credit Karma. Again from 10-K “A significant number of consumers that apply for a loan on learn about and access through the website of a loan aggregator, Credit Karma. The percentage of loan originations that were derived from traffic from Credit Karma was 38%, 38% and 52% in 2018, 2019 and 2020, respectively, and the percentage of loan originations that were derived from direct mail was 28%, 23% and 12%, in 2018, 2019 and 2020,respectively.”

“In November 2020, we experienced a reduction in the number of loan applicants directed to the Upstart platform by Credit Karma and a corresponding decrease in the number of loans originated on our platform, and we may experience additional reductions in traffic from Credit Karma in the future. If traffic from Credit Karma decreases again in the future as a result of this program or for other reasons, our loan originations and results of operations would be adversely affected.”

52% of loan originations from Credit Karma in 2020 (Credit Karma was acquired by Intuit in Dec 2020) is also a potential risk and the fact that Credit Karma can decide to reduce the traffic to Upstart platform can also hurt their revenues at least until they expand their platform to other customers. If they integrate Prodigy and open up their platform for auto-loans, they can have potentially unlimited avenues for growth.

Although I like what Upstart has to offer disrupting the lending industry, I think it is prudent to keep in mind of these risks. Any comments/insights on this will be much appreciated.



My take on this is the real risk is if Credit Karma creates a competing product and drops Upstart. Currently, they both benefit as Credit Karma receiving a bounty for referred customers, it’s a significant source of revenue. They would also take a hit if the customer flow slows down and that’s not in their interest. At this stage, this scenario is highly unlikely if Upstart’s AI is as game-changing as I think it is.

I wouldn’t consider this customer concentration. It’s marketing concentration and it’s very common, although still problematic. Think of how many businesses rely on Google for most of their traffic? It’s basically the same issue. And it’s always the number one priority of the business to develop new channels to get off of Google. The problem is that there aren’t always alternatives at the same scale. Hence why Google continues to face charges of being a monopoly.

As for the bank, they can probably buy one at some point. SoFi just did that. Finding new banking partners I don’t think is all that challenging. Interest rates are much more an issue here and they are controlled by the Fed, not the local bank.

I would file both of these issues as things to watch in the future. But I don’t find them concerning and think they are normal at this stage given Upstart’s relatively small size.


Nice discussion Babu and GolfCaddy. Thanks for your input and ideas. Much appreciated.

I think that Upstart’s AI-powered lending platform which incorporates more than 1000 variables not only gives them more variables but gives them a powerful moat. They get to use all their past outcomes to improve their accuracy. I remember them saying something like “every loan payment or non-payment improves our accuracy for the future, and we have an eight-year head start over any potential competitor”. It reminds me, in a way, of Crowdstrike with its knowledgeable of everything that has ever been tried to breach any of its clients, in the sense that both companies just get smarter and smarter over time.

Thanks again,



Good to see a discussion on UPST. Something I would like to ask is why do “we” believe they have guided to such huge acceleration in their revenues for this year. I didn’t find their explanation very convincing during their Q4 conf call, essentially catching up from the Covid slowdown. However previous FY growth rates were slowing and averaged around 65% and not 115%+. Could it be such a huge acceleration is down to there AI + ML platform becoming so smart (learning from 7 years of data) that is out competes other approaches ? I am guessing this is what we should expect from AI+ML over time and if correct then UPST is in a great position to deliver hyper growth for an extended timeframe.

Would appreciate feedback from any AI+ML experts. TIA.


“AI/ML expert” here. First of all, I always look askance at claims that revolve around a number of data points (“over 1000 signals”). This is marketing copy that tells you very little about their capabilities. You can throw any old data point into a model; it doesn’t mean it’s contributing useful information. At the same time, there’s no standard definition of a “signal”; you can carve up your dataset however you like to create an arbitrarily large number. Likewise, I don’t see Upstart’s head start as that significant. Larger institutions have massive customer bases to build models on.

I do see value in what Upstart (and similarly, Lemonade) are doing. It is a big cultural and process shift to move away from expert-created heuristics like a FICO score to a trained model. It’s not technically that amazing, so legacy institutions could replicate it. The question is how much inertia they would have to overcome to do so. It would surprise me if they would leave significant gains in efficiency on the table, but some places can be very old-school.

This is where I see a strong potential for Upstart (and not for Lemonade unless they pivot). These incumbents will find it much easier to buy these AI-driven efficiency gains than to build them. If Upstart can provide the engine that helps 50% of the country’s lenders write better loans, they could have a very sticky SaaS offering. As long as Lemonade is trying to actually be the insurance provider, I don’t see whatever marginal improvements their AI can make to their profitability as justifying a P/S multiple so far out of line with the industry.

I do believe that these AI approaches are the future of risk assessment in insurance and lending. But we shouldn’t underestimate the extent to which existing institutions with billions of dollars at stake are already trying this, and we shouldn’t overestimate the improvements they bring. If the story for these newcomers is just that the AI leads to better margins, it’s going to take a long time for that advantage to naturally outcompete the incumbents, and it’s likely to be swamped by other factors. Licensing the capabilities is the way to succeed, IMO.


My best guess is the accelerating economy and improving the job market will increase the pool of candidates that qualify for their loans. This is an unsexy but very obvious answer.

It’s interesting to note they charge more than competitors. The AI allows them to extend credit to people that traditional banks would overlook given their manual approval process. And they can approve people faster. This would indicate a significant advantage over competitors. Perhaps the speed to approval is allowing them to capture market share from weaker competitors who have no hope of developing their own AI. That’s a better story, but I have no idea if it’s true.

Read the Nerd Wallet review here for more.…


Upstart let me think about Ant, yes a China financial company which belongs Alibaba group. The Ant’s success experience is base on the mega consumer trading data from Alibaba Taobo or Tienmart. UPST don’t have something like this so I believe they may need some bank group feeding them the data to let them create what they called “AI”. I have to admit much risk from this model but it’s necessary because here is not China. You don’t have any privileges to do such kinds of business like Ant.

So I agree they need to lower their single customer dependency. I don’t know when, but looking forward to. Maybe they’ll be acquired someday by E commence.

Thanks Softie & GolfCaddy4PLynch. I did think the 1000+ datapoints seemed incredibly high (and surely the majority would have limited impact on the decision) when compared to what a traditional approach would use/ask for. However from what I have read and seen of Paul Gu he seems super smart and he and his team are responsible for the platform.

Will be interesting to see what they have to say tomorrow.

Coming from the financial industry, I think what may not be apparent to many people is how significantly the “traditional” banks are investing in AI. I think the biggest competitor for Upstart is the banks themselves.

Back in 2018, TD Bank acquired Layer 6, an AI startup to internally do what Upstart is doing plus a whole lot more. On top of that, think about how much data a financial institution has about you internally, that they are not sharing externally to companies like Upstart.

To read a little more about this:…

And I’m sure TD Bank isn’t alone, it is quite common knowledge within the industry that it is evolving towards AI:…

So where does this leave us with Upstart? I think there are 2 likely outcomes: (1) The company gets acquired, similar to the example above. (2) The company targets mid/smaller sized banks that don’t want to develop their own internal AI capabilities due to cost.

This likely limits the TAM, but doesn’t mean that it can’t be a very successful and profitable company.

Daws (long UPST)


So where does this leave us with Upstart? I think there are 2 likely outcomes: (1) The company gets acquired, similar to the example above. (2) The company targets mid/smaller sized banks that don’t want to develop their own internal AI capabilities due to cost


Isn’t it true that UPST by aggregating info from many banking sources would be able to develop a superior AI model, thereby becoming the go to source for good underwriting and decision making. Its model would be prefferred to one developed by any single institution.???



@draj “Isn’t it true that UPST by aggregating info from many banking sources would be able to develop a superior AI model, thereby becoming the go to source for good underwriting and decision making. Its model would be prefferred to one developed by any single institution.???”

Not necessarily. I don’t think the model would be materially superior. And, I think where the AI really shines is when a bank can internally develop a set of models that focus on predictive customer behaviour. The models will know when you are most likely to want to buy a home, and therefore can target mortgage offers to right people at the right time. This has nothing to do with making a better underwriting decision, but it has everything to do with increasing mortgage sales as a whole. So I believe the larger institutions are thinking much bigger than “Can I build/buy a better underwriting decision making model”. They are looking at the whole customer journey and using AI to predict what banking product you’ll want next.

At this point we’re probably getting off topic for Saul’s board, so I’ll leave it there. I did want to provide a viewpoint on where Upstart’s “competition” is and how it may impact their target market.

Please don’t take this view as “anti-Upstart”. I think this is still a very viable and valuable model for the smaller mid-sized banking segment that doesn’t have the resources to invest in its internal AI.

Daws (long UPST)


“Coming from the financial industry, I think what may not be apparent to many people is how significantly the “traditional” banks are investing in AI. I think the biggest competitor for Upstart is the banks themselves.”

Coming from the tech industry, I’ll take my chances with Upstart’s Dev team over all the banking companies in the world throwing billions at the problem. Ultimately, banks will use whatever tools bring them competitive advantage, even if they spent a fortune developing something in-house that doesn’t work as well.


Here are my few cents.

UPST claim reduction of credit losses by up to 75%. They focus on consumer loans for people with low credit scores - a segment that generally has very high delinquency rates - I’m talking 20% plus and loss ratios that are in 5% plus ranges.

IF… and this is a big IF, you can improve that loss number by 75% you go from losing 5%+ of the money you lent to 1.25%+. This is absolutely incredible and improves profitability of these loans DRAMATICALLY.

75% seems like a lot to me but even if you make that 25%, it is certainly enough to make this product more than worth its money.

Delinquency rates last year dropped dramatically, contrary to what I would have initially thought. This is mostly because a) people had less to spend money on and b) government handouts, loan concessions, repayment and eviction holidays. Given that these factors are going away with the pandemic ending, it is reasonable to expect that delinquency rates will go up substantially in the coming months. As a result, and regardless, it is far more important today than ever before for lending institutions to pay attention to more than their set of traditional factors they consider in making credit decisions, especially to this particular market segment - a segment that offers the most juicy returns if gotten right (ie getting low loss ratios). I suspect this may be behind UPST’s stellar guidance. They may not just think what I’m thinking, they may actually see it in their pipeline.

Re UPST’s dependence on one bank - this bank acts as an underwriter of loans that are sold off to other buyers. I suspect it would be very easy to replace them with somebody else who would be willing to do it for a fee.


Hi Softie, do no underestimate what data large banks actually have (especially historically), how incredibly arcane their systems are and how impossible it may be for them to recreate history to figure out which signals are relevant. Just look at the traditional credit bureaus. If somebody should be in the AI business it would be them but their credit models are antiquated and totally dumb in my opinion. They are so bad that their ability to predict defaults is no better than throwing darts. If somebody buys UPST, it will be one of them… that is if they ever realize that UPST is their direct competition…


Large banks have been in business for many years, they have not figured out how to use their massive data, how we can believe they will do so with AI now.

1 Like

Despite the recent haircut UPST remains my largest holding. Fortunate to have staked out a position in December, less fortunate to have added after the blowout quarter.

Quarterly Earnings report coming out tomorrow. They had better beat in a big way. If they are as smart as we think they are, they will not disappoint. No rational reason for them to have bumped up estimates so much unless they were sure they could blow out the number again, especially since 80% of Q1 was over when they issued the new guidance. Remember, it was March 17, when they reported 12/31/2020 and raised guidance.

In light of that, they likely need to report $140 million or so in revenue (vs $115 estimated) to please the market.

Regardless, i’m a believer in the longer term story, their machine learning lead, the associated flywheel effect, and in the first rate management team.

I don’t believe competitors can simply throw money and other resources at this to catch up quickly.
For this kind of business machine learning takes time, meaning years, for loans to process, for periodic payments and defaults to occur to sharpen the system, reduce loan risks, and thus more accurately determine interest charges to borrowers.

And even if UPST’s business remains centered around small and mid sized lenders, the TAM is still unlimited as the company learns to apply its technology beyond unsecured and auto loans, to include appliance loans, mortgages, and more.


Some thoughts on the UPST discussion, which had some interesting points throughout.

  1. I think large legacy banks may develop their own in-house AI product, but they won’t be sharing it with small and mid-sized banks. The latter are actually Upstart’s primary target market because these smaller institutions won’t have the resources to develop their own AI.

However, I don’t know if legacy banks (or even Credit Karma) will be able to attract the same caliber of talent as a smaller tech-based company like Upstart, potentially leading to a lower quality AI product (especially given they are 8 years behind at this point). Top engineers and developers typically prefer to work for startups or companies with significant growth potential (that also offer a lot of stock-based compensation). What’s more, these legacy banks are slow movers that aren’t very innovative, so Upstart’s lead will just continue to grow and these banks may have no choice but to partner with them or get left behind (look at what Square and PayPal are doing in the fintech space, taking significant market share from banks, especially among younger consumers).

  1. In addition to customer concentration risk, I am concerned about future lending patterns. As rates begin to increase from the current historic lows (eventually), there may be fewer borrowers overall which would, in turn, lead to less business for banks and fewer fees collected by Upstart. This would obviously lead to deceleration in their revenue growth. We know what happens when the market begins to see that trend. So the interest rate fluctuation potentially make UPST revenues rather cyclical over time.

Anyway, I like the concept behind Upstart’s product (particularly the fact that it might lead to more equitable, less discriminatory lending practices) and have a small starter position to keep it on my radar, though I do have the above concerns.
Also, as others have mentioned, the “1600 data point” claim that they always highlight seems gimmicky and I’m not sure what that materially means or if that number is
even significantt (e.g., vs. a hypothetical 500 data point AI model…when do the additional data points become statistically insignificant?). Thanks for the the interesting discussion…

  • Ceez

My first post here as a rookie investor. Thanks for this wonderful board.

I found this article about Cross River Bank. It’s from december 2019, so it might not be accurate anymore, but I thought it was interesting to post here.…

“Cross River is on a lending tear. It is underwriting loans at the rate of more than $1 billion a month—some $30 billion worth in just nine years. But unlike in banks of yesteryear, virtually all Cross River’s lending officers aren’t human beings. They are apps. Cross River’s loans originate mostly from 15 or so buzzy venture-capital-backed financial technology startups, so-called fintechs, that go by names like Affirm, Best Egg, Upgrade, Upstart and LendingUSA. The fintechs provide the customers; Cross River provides the licenses and infrastructure. It holds 10% to 20% of each loan it issues, and the massive volume of fintech loans has propelled Cross River to $2 billion in assets, up from $100 million a decade ago”

My 2cents, since the customers originate from Upstart and Cross River takes 10 tot 20% of each loan, there is not much chance they will suddenly end this collaboration.

  • Tim