UPST/Associated Bank vs competitors, n=1 sample

Using the link from yesterday’s press release of UPST partnering with Associated Bank for personal lending, I decided to try it out and compare it to other lenders myself.…

It is going to be an anecdotal, n=1 sample size, but I figured it would be fun to see the results nonetheless.

My FICO score is over 800. I have never missed a payment on anything before. My income is high.
Going into this experiment, I expected every lender to offer very similar interest rates within a 2-3% range, because I assumed that my default risk should be far lower than their usual borrowers (Spoiler, the results were quite surprising, with a huge spread of 19 points in APR)

I am not a previous customer of any products from Upstart or any of the lenders below.
I have never before checked a rate from Upstart or any of the other lenders below.
I am not a customer of Associated Bank or have done any business with them before.
No hard pulls were done by any lender/underwriter, so these are all “pre-approval” rates.

Associated Bank’s maximum offered personal loan is $25000, and I chose that as the requested loan amount.
I selected the loan purpose of “other” for every application.
Every input (income, address, monthly mortgage expense if asked, etc) for each application was identical.
Each application was entered on a mobile phone.
I did not select “yes” to co-borrowers if the option was offered.
Upstart did not require social security number input, but a couple of the other lenders required it for pre-approval rate checking.

Other details to note:
SoFi did require I input the monthly payment amount I desired, so I entered $509/mo to match the offered 5 year monthly payment plan by Upstart ($508.56). And SoFi did not offer a 3 year loan to compare.
BestEgg (Marlette) was the only underwriter that had you choose if you would do “autopay” or not, so I did select “yes” to autopay for their application.

The results for a $25000 unsecured loan request: Upstart offered the lowest APR out of everybody.

Upgrade: “For some reason we can’t offer you a personal loan today, but here’s a credit line of $5000 at 15% APR instead”

Upstart: 5 year APR 8.14%, no origination fee, monthly payment $508.56

SoFi: 5 year APR 9.93% (lowered to 9.43% “with 0.50% discount of autopay and SoFi Money discount”), no origination fee, monthly payment $524.19

BestEgg: 5 year APR 9.99%, origination fee of $1247.50, monthly payment $504.52

Prosper: 5 year APR 13.67%, origination fee 2.41%-5% (per website), monthly payment $548.56

LendingPoint: 5 year APR 14.99%, origination fee 0-6% may apply, monthly payment $594.58

Avant: 3 year APR (no 5 year option offered) 22.95%, no origination fee, monthly payment $967.12

LendingClub: 5 year APR 27.1%, origination fee $1500, monthly payment $719.06

KBRA data on average FICO scores of most recent securitized trusts for each lender:
Sofi FICO weighted average: 753
Marlette (Best Egg) FICO weighted average: 724
Prosper FICO weighted average: 715
Lending Club FICO Weighted average: 713
Upgrade FICO weighted average: 698
Upstart FICO weighted average: 674
LendingPoint FICO weighted average: 671
Avant FICO weighted average: 650

It is very interesting to see that SoFi and BestEgg (Marlette) offered loan APR below 10% while the other lower-FICO score lenders could not (except UPST).
SoFi and BestEgg very clearly favor only the most ‘prime’ borrowers and reject almost all of the below-prime folks.

We can also see this confirmed, from KBRA CNL data ( where Marlette (MFT) has base case loss projections of 9-10% and SoFi (SCLP) at 5-8%, implying a very high FICO score loan pool.
Meanwhile for Upstart, UPST had initial KBRA projections (at closing) of 14-22%.

Yet despite SoFi and Marlette’s hyper-focus on the best prime customer market only, their underwriting still could not provide me a better 5 year loan APR than Upstart (whose focus is clearly aimed at the below prime market, and so presumably Upstart’s data sets for high FICO folks like me would be limited!)

I therefore believe, for this n=1 sample, Upstart’s alternative data modeling was still able to beat out its competitors who had a richer data history of super-prime borrowers.


johnwayne excellent insight, thanks. Upstart is one of my top conviction stocks, and they do things very differently from their competitors as is clear from what you posted.

What you posted got me thinking about another n=1 anecdote which I thought to post.

I am based in Europe, so not a prospect for any of the companies listed above.

Over the past months, when reviewing Upstart’s site, I noticed that I was unable to access certain parts of the site, so wondered whether this was also unique to them and could possibly add to their data advantage.

So today I decided to do the same as you and apply for $25k loans on Upstart’s and all of the competitor sites you list above, but from Europe and with reasonable, but bogus details, also on my mobile phone.

The results are as follows:

Upstart - Before being allowed to view the site at all, I was asked to click that I am human and then to select pictures of a truck to prove it. Then, when I tried to check my rate, I was stopped from even starting the application process. On the home page just under “Get a smarter loan, what would you like to do” I clicked on “pay off credit card” and then got a big “access denied” screen. I could view everything on the site but the moment I tried to apply for a loan I was stopped.

→ Upstart therefore stopped junk data at two points: first it confirmed that I was a human before I started browsing the site, ensuring that the browsing data is legit, and then, even though I am a human, it stopped me from trying to apply for a loan at all because, I assume, me being in Europe meant that I am not a prospect. I asked friends of mine in South Africa to try the same and they got the same result: access denied to the loan application process.

SoFi - stopped me at the beginning of the process by sending a message to my bogus phone number.

→ SoFi also stopped a full bogus application from making it into their data of sign-ups.

Avant, Lendingpoint - accepted all of my bogus details including account creation with a password and then told me that unfortunately they cannot provide me with a loan at this time, and they will contact me via email with more info.

Prosper - accepted all of my bogus details and then presented me with a message “Credit file not found. TransUnion does not have a credit file on record for you and could not provide us with a valid credit score”

Lendingclub, Best Egg - accepted all of my bogus details and then Lendingclub presented me with a screen titled “Unfortunately your loan request couldn’t be approved” with a formal-looking Adverse Action Notice addressed to bogus me. Best Egg’s was similar.

→ Only Upstart prevented junk data from polluting their sign-ups in my case. SoFi prevented a full application from being recorded but will still record a bogus partial sign-up and subsequent drop-off. All of the others have a full junk sign-up and account polluting their data.

That may be completely irrelevant to them as all of them prevented offering bogus me a loan or rate, which could be all they care about. However, assuming that there are more people doing junk sign-ups or test sign-ups or right-out fraudulent ones out there in the world (as there surely must be) they cannot use their data to learn before cleaning the data of junk first.

Upstart does not have my junk sign-up in their data at all, as they stopped me from going through the application process before I even started, while none of their competitors did.

I do not know how Upstart prevented me from applying/checking my rate (I assume it’s based on my IP address) but the fact that they did means something to me: their data is cleaner than their competitors’, meaning that they can improve and learn from it better than them. If, as I’m sure is the case, they have many more of these types of processes stopping garbage data in their site (Paul Gu said as much in an interview), then their data is much, much cleaner than their competitors’, more relevant, and getting more so with every sign-up.



Hi wsm007.
Like you I live outside the US and even accessing without a VPN forces me to prove that I’m not a robot by battling through that Captcha nonsense.
This is particularly annoying since:
a) I can’t discern whether a kayak or a canoe can be classified as a boat.
b) Whether a traffic light pole counts as part of the traffic light
c) What on earth part of a motorcycle is identifiable by a close up of a red tyre.
d) I’m a robot.

But having said that then at least I can go directly to their investor relations sections via a Google search after I emailed Upstart to complain that their investors from outside the USA might actually want to listen to their earnings conferences.

So… I wouldn’t read too much into their Captcha process weeding out anyone other than by your (and my) IP address.

But anyway, to add some value to this post then I’d highly recommend this Youtube interview with one of their Machine Learning team. I think that he explains why what they are doing is unique:

Cheers, PB.


Great analysis!
Lower loss rates could mean better underwriting standards and vice-versa!


One point I did not emphasize enough in my post:

I believe one major reason why the APR offered to my “superprime” status is so high from the lower-prime lenders/underwriters (LendingPoint, Avant etc) and the APR offered from high-prime lenders (BestEgg, SoFi) is lower, is that higher losses incurred by low prime lenders must be made up for by charging everyone else higher APR to cover the costs of defaults and volatility.

UPST cofounder Paul Gu explains this concept well from a 2015 talk (when he was like 23 years old!):

So, UPST can offer lower APR to everyone on the entire FICO spectrum than its competitors, despite primarily lending to low-prime folks, because UPST’s superior AI underwriting is avoiding traditionally expected lower-prime default losses.
Savings from lower defaults are passed around to all.

Also, I dug up some more findings showing just how rigid other lenders’ underwriting models are. LendingClub and Prosper have been pricing unsecured loans with very traditional underwriting methods for years.

Here was some analysis by a PhD from May 2019 (I just looked him up, the author, coincidentally, currently works for Upstart in their machine learning department since being awarded his PhD in Finance):…

“The mean FICO score of Lending Club and Prosper borrowers is in the 700–710 range, with the borrowers being tightly concentrated around this mean. Subprime borrowers are automatically rejected and borrowers with very high credit scores rarely apply…
…Overall, the summary statistics support the notion that the two lenders are direct competitors. They seem to be offering basically the same products to the same demographic.”

This is further evidence that at least LendingClub and Prosper have been doing nothing special vs UPST’s AI/ML. They are subscribing heavily to the idea of FICO score = ‘everything’.

Meanwhile, UPST’s senior VP of biz dev Jeff Keltner has stated in one of his podcasts that if UPST encountered a borrower who didn’t have a FICO score, their AI model would likely still function all the same, by using their other 1600 alternative data variables.