UPST-Kemba webinar. GREAT insights

I finally had some time to watch a Kemba credit union webinar with UPST that was recorded on June 15, 2021. I came across it after reading a NAFCU press release from last week:…
The webinar, titled ‘CU Case Study: Growing Your Consumer Loan Portfolio and Gaining New Members’, involves the VP of Credit Administration at Kemba answering questions provided by UPST’s VP of Lending Partnerships.…

This was one of the MOST insightful sources I have come across lately regarding UPST partnerships. The key takeaways I had are at the very bottom of the post. I know this post is very long but for any UPST holders on this board, I guarantee it is worth reading through each word. (Or you can watch the webinar yourself too) Here are the notes I took:

How can credit unions manage the current environment?
There’s been a huge surge of deposits. From 2009 to 2015, credit unions averaged 6-8% deposit growth a year, but there was up to 20% deposit growth since 2020 (double or triple the growth average).
Historically low interest/low yield environment also makes it difficult for credit unions to cover costs.
Mortgages have been red-hot, but overall consumer loan demand has downtrended since pandemic.
And more credit union members demand a digital experience.

Background on Kemba
The Kemba executive on the webinar says he is the VP of Credit Administration: a new role since 2021 separating back office and underwriting function from sales. He was previously the equivalent of VP Consumer Lending before 2021 at Kemba.

Kemba Credit Union was founded in 1933, has assets of $1.8 billion, with 100,000 members in Ohio. Kemba is an Upstart partner since December 2019, and joined the Upstart referral network in February 2021 (which the referral network became generally available to all other partners, in March 2021)

Kemba has grown its consumer loan portfolio and acquired new members via UPST. Kemba uses UPST’s referral network. UPST does all frontend work and all the marketing and the loan gets to Kemba in the end.
A soft credit pull on UPST website immediately gives an offer within seconds via AI, and 70% is automated verification.
Then the applicant signs a Kemba loan agreement AND a Kemba membership. Once per day we can send out the ACH deposit to all new members. So it’s a same business day loan.
All Kemba has to do is board the loan onto Kemba’s IT system. All this happens same day.
We are not needing to lift a finger for any part of process except boarding.

Poll to audience: What is your credit union’s greatest challenge in 2021?
75% Growing our consumer loan portfolio/putting cash to work
12.5% Member acquisition and retention
0% Expanding into new markets and/or products
12.5% Increase digital capabilities

It’s clear most credit unions on this webinar have the same challenges. For us we noticed there were big loans on credit reports of Kemba members, that they had unsecured loans from fintech lenders and not from Kemba, they never even bothered to apply to Kemba - why? We thought we were already knocking it out of the park on our unsecured internal product, that if members physically came to our branch we could service them well. What are fintechs doing right that we thought we offered to our members better? We even had lower/better rates than the fintech lenders. Well, we noticed members were expecting speed/availability – choosing that convenience even over lower loan costs/better rates. Netflix, Amazon has trained consumers in this demand expectation. So we asked ourselves how do we get into the AI or fintech space, we want to do this but just have no plan. We HAVE to do something, there has to be movement on this.
The average credit union member age is 47 years old. On average, credit unions across the nation are aging. Many of our older members love to come into the branch. But how can we replenish our Kemba members with younger folks who are more likely to demand more credit services than the older members? There’s an opportunity here to attract demographic that wants to go digital. There’s a NEED. If we can’t service them, others will.

Why did Kemba consider the personal loan to make the leap and grow this specific portfolio?
Like most credit unions, our auto loans were our bread butter. We then learned from indirect auto lending [what’s indirect lending? A lender makes a new loan to someone that they did NOT have a previously existing relationship; the dealership selected the lender to make a loan at point of sale to a new borrower] that people were coming in to get the loan, but not converting over to other services. We want members who want to be part of our community and not just for a single loan. We saw members survey scores weren’t feeling too great about coming in via indirect lending. They were searching to buy a car, so after getting an auto loan, Kemba was just secondary to them; but a personal loan was different since they were shopping for cash; if a person wanted cash that’s very very different, and if Kemba gave them cash they were more likely to see us as a financial partner. A personal loan is more about solving a need and it might be the one thing that gives them a good night’s sleep. It builds trust, builds loyalty that credit unions are known for. Personal loans let’s them consolidate bills, cover a big expense like dental work. It’s a flexible product. We still offer a similar unsecured product internally but now UPST is giving us volume equal to our internal unsecured product, so it has more than doubled our existing volume in the first year, and we noticed that 5-10% of those coming from the UPST referral network is coming from EXISTING members – these are existing Kemba members who didn’t even choose to apply for our internal unsecured loan product, they went straight to UPST! They didn’t come to us first – so clearly, there was a digital need that we were missing the boat on, that we just didn’t have that better solution that UPST offers. There’s a specific group of people that like our process for internal unsecured loans, but clearly there’s a whole demographic we were previously missing out on.

We bring members through that indirect channel by making calls to them. We found that the phone call to those folks who bought a car is far different than those who came through Upstart for a personal loan. When we say ‘hey I sent you the money yesterday, did you get that’- it’s a far better call for our associate and the member on the call appreciates it more. We gave what they needed same day, and we verified it the next day. It doesn’t line up too well with the indirect auto car buyer who might get a call from us literally a few weeks after the purchase happened, versus calling our new personal loan member immediately.

From Kemba’s standpoint, again we thought we were doing a great job and working really hard on our existing personal loan product, but after partnering with UPST for a few months we realized we really weren’t capitalizing anywhere near our potential. Getting a specialist that specializes in doing a certain product opened our eyes.

Poll to audience: Where is your personal loan business today?
0% We don’t offer them and don’t buy them
0% We don’t offer them but we do buy
75%We offer them, but have a small portfolio
25% We have a large and growing portfolio

I noticed no one selected the ‘buy portfolio’ option. What’s your take on originating the loans versus buying them?
I do believe from rules/admin standpoint we have some restrictions, we do some participation loans with other credit unions, but we believe we offer a good service and getting that member onboarded to Kemba through our website or branch or virtual branch we can provide that service, having that member relationship is more important to us than buying that loan.

How did you make the decision to build it on your own versus partner with someone for personal lending?
Our first thought was let’s try and build this ourselves and figure it out. We were proud of ourselves because we got to 50% automation on our product, but when we wanted to add this variable or that variable we quickly found out our system couldn’t make it work, or if we were able to add a variable then the compliance department would come over and say ‘no you can’t, you need to add a red flag on that.’ We worked on it forever and couldn’t get it. We were struggling. But UPST could get to 70% automated of approved loans while we were barely getting to 50% automated, and that was with including loans that were automatically declined! So we moved on to our existing online vendor and tried to set up unsecured loans but then we found out this vendor wouldn’t work with our LOS (loan origination software), or this other vendor wouldn’t work with unsecured loans, or this one vendor doesn’t partner with our core system. Nothing could get us across that line, but we thought let’s still try to find a vendor that is a one shop stop. Then our board also started asking ‘hey what about AI, what about fintech, did you consider that?’ We realized we were only a $1.8B asset credit union offering all types of consumer products, versus a partner out there who has teams of data scientists focused on one specific loan type. UPST was just so much further along than we ever could accomplish in our own process.

We were initially pushing our underwriters to get loan decisions to 10 minutes but UPST is instantaneous. We don’t have access to verification processes and checks that UPST does. We were also nervous about fraud; how aggressive do we get to verify? it’s not ‘give us a paystub and we get back to you in 2 days.’ UPST is doing it so fast, members don’t realize verification is happening at <0.3% fraud rate. Our members were coming back, and NPS service scores are through the roof, the Upstart scores are higher than our own internal member score averages. Members are telling us that ‘we’ are doing it right with UPST.

What do you look for in a partner? How do you determine a partner to engage with?
First, our goal was to help more people. We felt that 700 or above FICO folks, they had options. We were looking for a vendor targeting the below 700 FICO people that would still pay the loan back. We want a partner who was responsible. We didn’t want partners that wanted to charge 30+% and be like ok we have all these charge-offs but it’s fine because we can still make some net money.
Second, was technology. We didn’t want a company that would fight against us over loans, that is, we don’t want a fintech competitor to partner with. We found UPST was a tech company. Loans were secondary to them. They were trying to be responsible, they were looking for that right member who would be responsible, finding that true risk, and we had room to grow in the partner relationship, it wasn’t that we had to be ‘all in’ right from the start, we had lots of time to ease in and understand what UPST was doing first, through the referral network - which made all of us very comfortable.

How would our listeners think about talking about this internally with executive teams? How to get their leadership comfortable?
First, get over examiner fears. We felt initially “there’s not a lot of data here, we don’t have experience.” But that UPST No Action Letter got us comfortable to approach the examiners [The NCUA is a federal government agency. From credit union examiners plan, conduct, and complete audits of federally chartered credit unions and the workday varies from one credit union examination to the next. Examiners play an important role in maintaining confidence in the credit union system by ensuring that is operating in a safe and sound manner and complying with all federal laws and regulations.]
Second, we thought possible disparate pricing was going to get us in trouble versus risk based pricing but we went straight to our examiners and laid it out, they gave us support.
Third, being a lender, we feared not having control or full understanding of the thousands of variables involved in the AI underwriting decision. This part just simply took time for us to be comfortable.

Poll to audience: How has your credit union responded to consumer lending since the pandemic began?
4.5% We have raised prices/rates.
95.5% We have maintained our consumer lending program levels.
0% We have stopped some or all of our consumer lending.
0% We have slowed down consumer lending.
0% We have raised credit score requirements.
0% Other

This is not unexpected to see; despite uncertainty of pandemic, loan demand has fallen too. It’s hard to make a strategic decision change.
At the beginning of the pandemic we were nervous about unsecured loans internally, and we cut back there, but internally we do risk based pricing on FICO. With our weekly calls with UPST we found out UPST puts in economic factors into their AI which our rigid risk based pricing couldn’t do. We didn’t know what the hidden risk was in the pandemic but we appreciated what UPST could do and gave an advantage.

How did Kemba’s program with UPST perform?
For loans we got through UPST, in the first year, we had only one loan that went bad. And our charge-off ratio is 14 basis points (0.14%). Well beyond our projections and expectations. And our member care scores are through the roof.

I’m seeing questions from folks about the referral network. What controls do you have?
Monthly, we can set the loan amount, volume, debt to income ratio, etc. In the beginning we went through each one line by line to figure it out. Then we got comfortable - it’s AI for a reason. We should let models do what it should be doing. But we don’t want charge offs to be 10-12%, we want responsible members, so that’s really the only thing we were focusing our control on.

Regarding question on geofencing and servicing.
Only members that meet Kemba’s geographic area are fed through the Referral network to Kemba.

UPST can service loans on behalf of Kemba, but Kemba has the resources to choose to service.
But what options are for credit unions that aren’t like Kemba? We at Kemba are interested in a CUSO [What is a CUSO? Credit Union Service Organizations operate as back-end, profit centers for credit unions to help them pool resources and reduce service costs. CUSOs must have at least one credit union owner and most CUSOs have a specialty, such as auto loans, technology resources, commercial lending or consulting services.] This relationship with UPST is a partnership to potential other product offerings down the road.
Lots of people have questions about UPST in general and anyone can come to me with questions - my contact is on the side, and in the slides at the end. We’re definitely interested in forming a CUSO with other credit unions to get others on board with UPST and spread the word.
A CUSO allows multiple credit unions to work together with our due diligence to leverage things that Kemba has already established. Even though we do the servicing, we’re still really limited, the loans are boarded by us and we only have so many people available for this but if we pool this up if we join forces we can make this a much stronger team.

Now that you built this program for personal loans, what’s the next area of lending you want to focus on?
What I learned most from this process is you probably need to put out a new product every year. We’re currently working with UPST to put a white label on the unsecured lending, and offer it to our members.
UPST has a pilot program on auto lending and ‘other products’ that we’re interested in. The next few years it’s focusing on the digital channel and how to take it to other products. We had no idea the opportunities are out there until we got into this relationship with UPST.

Yes. Auto is exciting. We launched it last year. It’s interesting the benefits we provide in personal lending is tremendous in auto at 6 times the size for members.
Fast forward in 2-3 years. What changes do credit unions need to make to better serve members?
It’s looking at our partners and understand which partner gets us closest - it’s not just simply throwing things online but it’s about branching things out. UPST gets members sourced from all sorts of places versus just a physical branch where we can only convert that to an online process on a website.

Let’s recap: Kemba is partnering with UPST to grow loan balances and acquire new members. Remember, these are NOT loan purchases. These are loans originated on Kemba loan documents. Through use of AI lending, Kemba has acquired creditworthy members staying within loss targets. Kemba is still lending to people who pay back, just finding more of them. We’re engaging with Kemba on a modern digital experience. Credit unions have to adapt and meet the member on that same platform. Let’s take more questions from the audience.

Operator: How is Kemba onboarding new members coming through UPST?
When we get the loan, IT has made it so it’s boarded immediately. We use zip codes so automatically they will qualify. We send funds out at 4-5PM by ACH each day, then we do a phone call the next morning to verify it was received, and another call 2-3 weeks later to follow up. Instead of mailing with indirect auto loans, we now have emails for this that go out right away for responses too.

Operator: How does UPST work if we want to use it on our own credit union website for existing members?
So one way is their referral network where UPST drives all the marketing and matches those borrowers to credit unions and turns them into members. Then another way is the same underwriting process and models from that referral network that can be offered the credit union’s online portal website through a completely credit union branded experience. The key difference is referral network provides the marketing versus a fully white labeled product.

Operator: How does UPST identify potential members and when do the potential members see the credit union branding?
UPST VP answers: UPST uses email, mail, aggregator networks etc; casts a wide net across the nation. 60% are doing applying on their mobile phone in checking the soft pull rate. After a soft pull, if underwriting criteria and member qualifications are met, then they are immediately presented within seconds with seeing the credit union branded loan.

Operator: How does credit performance compare across your two loan programs, UPST referred versus internal unsecured loan?
We’re a year in, so a little difficult to fully compare, but we’re 14 basis point charge-off ratio versus 150-200 basis points internally. UPST loan amounts are also a lot bigger than what we typically lend. We don’t have any delinquent loans right now, UPST is doing a great job finding the right members.

Operator: How does UPST find someone who qualifies for membership?
Well, we use zip codes. There’s also a couple checks we do internally specifically to Kemba, but a little too granular for this webinar, so if anyone has questions send it my way.

Operator: I do see more questions rolling in but we’re at the time limit, and we’ll make sure we follow up with you.

[My take away points]

  1. What a bombshell!!! To me, this is HUGE. Kemba is so excited by UPST results that they want to form a CUSO to spread the word about UPST to other credit unions. I mean think about the fact that this Kemba VP spent all this time and effort on a webinar to then poach other credit unions and be open/available to anyone for questions. A credit union is doing the marketing for more partnerships for UPST. And they are eager to expand into any new UPST product offerings as quickly as possible. WOW! And there seems to be lots of existing interest from other credit unions who wanted to talk to Kemba about their UPST experience. This webinar was an exciting inside look that lends credence to UPST CEO’s statement that he will be surprised if UPST doesn’t have hundreds of partners in a couple years. [And here’s my wild guess as to why he might want to form the CUSO too; maybe if other credit unions also get onto UPST’s referral network, then Kemba can provide loan servicing for those other credit unions within their CUSO, and therefore give Kemba extra income?]

  2. Kemba joined UPST’s referral network in February 2021, and since then, months later, has DOUBLED their previously existing personal loan volume. Now, imagine what revenues UPST would generate almost instantly if they landed a huge bank partner hundreds of times the size of Kemba (like 200B to 500B in assets)… Of course, a factor to this, is Kemba lowering the monthly ‘control dials’ because they quickly became comfortable with the AI underwriting process. So I guess not ‘instantly’ but could be very ‘quick’ if the bank is big enough.

  3. There are lots of borrowers who would choose CONVENIENCE/SPEED over lower/better loan rates/prices! Existing Kemba members were choosing to go through UPST and not even bother applying through their credit union! This makes sense in terms of matching the Trustpilot reviews - where the most common type of comment was the quickness/ease of the loan app process. And, UPST’s net promoter scores were higher than Kemba’s internal product, which again supports that idea.

  4. Credit union members’ average age is 47 and aging over time. Credit unions NEED fresh blood and UPST offers that to them with the average borrower age at 28 and average income of $86000+. They know this, and UPST is an easy sell.

  5. It’s hard to build your own digital/fintech type loan offering, at the least for a $1.8B sized institution. As the Kemba VP said, so many problems trying to do it themselves – vendors couldn’t fit, compliance creates huge roadblocks. UPST has a huge regulatory moat and the CFPB NAL was specifically mentioned as easing that anxiety over compliance in AI lending.

  6. Even if a bank partner on-boards immediately with UPST, unfortunately, it still takes time to ramp up. It takes at least several months to a year for bank/credit unions to get comfortable enough to lower those monthly control dials and let the AI do its job. Building the bank partner ecosystem is a long term process and not an immediate ‘big’ revenue generator.

  7. This is a common theme I’ve seen from this and other UPST webinars plus the UPST Leaders in Lending podcasts. This topic is always discussed in each webinar: “how can we sell UPST to our bank/credit union leadership?”
    This webinar provides evidence for what UPST CFO said in a fireside chat (…), “The way banks start to work with us is they have P and L owners that immediately want to work with us, it’s always an easy conversation”. Credit union vice presidents of consumer lending almost instantly WANT to be with UPST, but it’s these credit union VPs trying to figure out how to sell them to their internal executive red tape/bureaucracy. That’s the bottleneck here and why these partnerships take so long to on-board.

Finally, a question to the board: Any math experts out there can do an estimation? One of the audience poll questions above had a 4.5% plus 95.5% response. Does this mean there were at least 200 credit unions participating in the webinar?? Seems too preposterous…but in order to come out with these decimals, wouldn’t you would need 9 people choosing 4.5% and 191 people choosing the 95.5% option? Is there any way to size that even number down to still match the percentages? Or am I way off the mark there? To me 200 seems unlikely, since I would have assumed at least a few would have chosen the other answer choices…


For example, 44 people could also get you to 4.5% (2 / 44) and 95.5% (42 / 44) if you round to 1 decimal.

But a very interesting webinar indeed. Thanks for posting!


Hi jon. Thanks so much for your post. It truly was worth reading every word.

In regards to your question, you need to take rounding into account. 4.5% could be 1/22 or 2/44 or 3/67.


Sorry to burst your bubble :frowning:


Thanks 5thhorseman and jvc87.

You are likely right. But one of the questions had a 12.5% split. To make it even, 12.5% of 22 or 44 is 2.75% or 5.5%. I suppose it’s possible two persons of the 22 or 44 participants did not answer that question, which is what produced that percentage.

1 Like

Excellent write - up jonwayne! You’ve upped the analysis chops on this board immensely. Thank you.

One item you did not highlight in the takeaways is that the charge offs in the upst portfolio are 1/10th (an order of magnitude!!!) lower than their traditional loan portfolio. Banks are absolutely registering this. It will sell itself.

Long upst


Math teacher here - I can verify the math of 5thhorseman and jonwayne.

Further, I went searching for some more statistics about this, and found some data from the Federal Reserve. The table below shows

(1) the total oustanding amount (in billions of dollars) in nonrevolving credit accounts. This “includes motor vehicle loans and all other loans not included in revolving credit, such as loans for mobile homes, education, boats, trailers, or vacations. These loans may be secured or unsecured.”

(2) the average interest rate (%) for 24-month personal loans

     2016   2017   2018   2019   2020   2020   2020   2020   2021   2021   2021   2021   2021
      tot    tot    tot    tot    tot    Q2     Q3     Q4     Q1     Q2     May    Jun    Jul

(1)  2676   2814   2953   3100   3210   3151   3180   3210   3257   3321   3302   3321   3333

(2)  9.69  10.13  10.32  10.32   9.51   9.50   9.26   9.65   9.46   9.58   9.58   n.a.   n.a.

Note: I have no idea why their data is missing Q1 for 2020.

Link to source:

The takeaway here is that consumer lending is (1) increasing and (2) becoming cheaper over time.

Kind regards,
Trey C.


That was an interesting read and thank you.

One of the things that stuck out for me was how determined Kemba was to build their own system initially. They ran into lots of struggles and concluded that they were just too small an operation. This sticks out to me because bigger banks do have more resources to build their own systems. Clearly Kemba is very happy with their UPST partnership and it does seem a positive partnership for a small credit union. It doesn’t though speak to UPST being as advantageous for a larger bank.

Here is the part I’m referring to:

Our first thought was let’s try and build this ourselves and figure it out. We were proud of ourselves because we got to 50% automation on our product, but when we wanted to add this variable or that variable we quickly found out our system couldn’t make it work, or if we were able to add a variable then the compliance department would come over and say ‘no you can’t, you need to add a red flag on that.’ We worked on it forever and couldn’t get it. We were struggling. But UPST could get to 70% automated of approved loans while we were barely getting to 50% automated, and that was with including loans that were automatically declined! So we moved on to our existing online vendor and tried to set up unsecured loans but then we found out this vendor wouldn’t work with our LOS (loan origination software), or this other vendor wouldn’t work with unsecured loans, or this one vendor doesn’t partner with our core system. Nothing could get us across that line, but we thought let’s still try to find a vendor that is a one shop stop. Then our board also started asking ‘hey what about AI, what about fintech, did you consider that?’ We realized we were only a $1.8B asset credit union offering all types of consumer products, versus a partner out there who has teams of data scientists focused on one specific loan type. UPST was just so much further along than we ever could accomplish in our own process


Hi Fanae,

The difficulties outlined by Kemba above did not strike as me as things related to their size. For example, when they say:

“but when we wanted to add this variable or that variable we quickly found out our system couldn’t make it work, or if we were able to add a variable then the compliance department would come over and say ‘no you can’t, you need to add a red flag on that.’ We worked on it forever and couldn’t get it.”

I don’t see them drawing this conclusion necessarily because of their size. System compatibility or issues with the compliance department are unrelated to the size of the Bank. Same with outside vendors systems not working with their LOS (loan origination software) or their core system. It’s possible that they couldn’t overcome these problems because their size did not allow them to bring the necessary resources to bear on the problems but they did not say so.

In the end, fixing these problems may be more time and expense for banks of any size than it’s worth when they could partner with UPST and avoid these issues.

Definitely something to dig more into to shed light on UPST’s possible first mover advantage. Thanks to all for the illuminating information and discussion.




Thanks for a great post JonWayne.

One thing that I have not been able to find is, how exactly does UPST make money? I’m assuming its an underwriting fee of sorts? Is it as simple as a set fee per loan?

I see on their Q2 financials that revenue was 194m and their “Bank partners originated 286,864 loans”. So doing simple math that averages out to $676.28 per loan.
However, for forecasting, I’d like to understand their fee structure. Is it a flat fee or a % or some other method.

Thanks in advance to anyone with answers or directions to an answer!


1 Like

JonWayne, I am humbled by the level of research and analysis that you have brought to this board with respect to Upstart. For example, I was astonished by your attempt to determine how many attendees were present on the webinar based on the poll statistics. It doesn’t particularly matter if you came up with the right number, it was simply the attempt to make that determination that brought me up short. I would have never even thought of doing that.

But, setting that aside, my confidence in this investment was boosted by the discussion of the difficulties that Kemba encountered as they tried to develop their own software. Adding variables demonstrated that their system was a house of cards that would collapse due to the complexity. Integration with other systems proved to be a nightmare. Even though my IT experience is very stale at this point, integration and regression testing was always something like a battlefield when bringing a new, complex mainline system into production. Apparently, the folks at Upstart have developed the APIs needed to quickly integrate with most of the commercial software that banks are likely to have installed. And then there’s the pushback from their internal compliance department. And the potential for other regulatory issues. They also cited their inability to get past a threshold of 50% automation. And I’m sure their were myriad other problems they encountered which just weren’t mentioned.

Anyone who might argue that Upstart doesn’t have a moat fails to comprehend what an accomplishment is represented by their software. One might try to argue that a big bank, a BofA or JPM Chase or CITI or you name it with barrels of money would be able to replicate the Upstart software and compete with them head to head. But this argument doesn’t recognize that the primary resource needed is not money. There’s a skills and knowledge deficit that has to be overcome. That resource is not readily available and even less easily assembled into a dedicated development team. It was possible at Upstart because the lead scientist was also a founder. How would a big bank compete with that?

And finally, when confronted with all the difficulties of trying to compete with Upstart one must ask why would they? Even if a Big Bank were able to surmount all the difficulties in order to have a competitive offering. It would take at least a few years to do so when they could simply partner with Upstart and bring the offering online in a few months. Upstart’s commission is not onerous. In fact, it’s quite modest. By partnering with Upstart banks not only avoid all the difficulties of developing the software, they also avoid the on-going need to maintain it and keep abreast of the ever changing state and federal regulatory environments while simultaneously extended the capability to address other segments of the loan business.

Upstart is not just disrupting the banking business. They are also disrupting the predatory payday lending business by giving the creditworthy unbanked a legitimate alternative.


A reply to tonyh28 who said:

"One thing that I have not been able to find is, how exactly does UPST make money? I’m assuming its an underwriting fee of sorts? Is it as simple as a set fee per loan?

I was able to find the following in Upstart’s S-1 filing on the SEC website (page 71):

"The below table summarizes the dollar value of our economics on an average-sized loan, based on our contractual rates that were in effect as of September 30, 2020.

                       Paid by               Fees per loan

Referral Fees       Bank partner        $400-500 on origination

Platform Fees       Bank partner        $200-300 on origination

                    Bank partner
Servicing Fees    or institutional        0.5% - 1% per year

Loans on our platform today are predominantly sourced from For these loans, we incur variable costs in the form of borrower acquisition costs and borrower verification and servicing costs; in the nine months ended September 30, 2020, this category of loans generated a 44% contribution margin on average. Borrower acquisition cost and borrower verification and servicing cost are highly correlated with the Number of Loans Transacted on our platform and trended upwards on an annual basis. A small number of loans were sourced directly through bank partners in which we received no referral fee and incurred no acquisition cost; in the nine months ended September 30, 2020 this category of loans generated a 67% contribution margin. In 2018, 2019 and the nine months ended September 30, 2020, the average contribution margin per loan of all Upstart-powered loans was 15%, 31% and 44%, respectively. The rising level of automation and continued improvements to our Conversion Rate achieved through our increasingly sophisticated risk models and our evolving channel mix have contributed to improving loan unit economics over time. We further believe that bank-sourced loans can be an important driver of volume growth in the medium-term future; to the extent we are able to increase the number of loans sourced directly through our bank partners, our contribution margin would be positively impacted.

That statement in bold really stands out to me.

Looking at Upstart’s website, I also see that banks can pay for access to their “Credit Decision API” - which seems to mean that they can use Upstart’s AI technology to make lending decisions for other credit-based products.

I hope this post was helpful to you.

Kind regards,

Trey C.


I don’t know enough about how fragmented and inefficient the insurance industry is in the US, but whilst Upstart picks off more and more of the lending decision and origination market, I can’t help wonder whether they have a role to play in use cases for insurance decision making and origination and insurance pricing with their AI tools, which seems to be ripe for AI/ML disruption and how the size of this addressable market compares with lending.……



They are already using AI to underwrite policies, per their website:

Underwriting at Lemonade

“Powered by tech, Lemonade is able to collect about 100x more data-points per customer than traditional insurers (whether online or through the app). Lemonade’s AI-powered bot is designed with underwriting algorithms in place, which means most customers are able to get insured instantly.”

R4M (long UPST, no position in LMND)