Note: I was tasked with researching this company for four days. The following is the result.

Company History

UPST was founded in early 2012 by Dave G., Anna C., and Paul Gu. It IPO’d in 2020 with a follow-on offering in April 2021. The company began selling ISA and tried to use data to predict individual’s future success rate. The idea didn’t scale and one year later, with six-months of cash left, they decided to leverage the potential of AI in another way. The story goes that Paul Gu – a 20-year-old, double-major Yale graduate, with a six-figure salary – went to an online lending platform only to be confronted with a 27% APR.

At this point, the company decided to pivot their focus to the consumer credit space, more specifically, the unsecured personal loan industry (“UPL”). The began by partnering solely with Cross River Bank for ~5 years, to help them make better credit decisions in the UPL industry. Whilst the typical bank may use 15-30 variables, a loan-officer, and a simple regression analysis, UPST used +1,000 data points. For example, whilst a bank may only consider whether someone has a college degree, UPST considers the college, the degree in which the person majored, the GPA, etc. Paul Gu has said that these variables “are not only correlated with outcome in credit, they actually cause outcomes in credit”. Fundamentally, the premise of UPST is that even “consumers with high credit scores (…) pay too much for loans because (…) they (…) effectively subsidize the losses from borrowers who default.” Educational factors are important in predicting a borrower’s behavior in the first few years after leaving college, where credit history is scarce.

Business Model Overview

Today they have 18 bank partners to whom they provide an AI lending model. The software/API can be incorporated into banks’ systems and they can set their own parameters (such as a minimum FICO score, borrower location). UPST did $230mn in revenue and almost $12mn in operating income in FY2020.

UPST makes money through fees. There are three kinds of fees: (1) referral fee (3-4% of loan principal amount), (2) platform fee ($200-$250) and, (3) servicing fee (0.5-1% of outstanding principal over loan life). UPST is entitled to referral fees when they refer a customer from their own website to a bank partner. Platform fees are fees banks pay for using their software. Lastly, servicing fees are fees paid to cover costs of reconciliation of payments received, borrower support, etc. Customers can borrow money using UPST in two ways: (1) they either go directly to UPST website who then refers them to one of their bank partners or, alternatively, directly through a bank where UPST’s system is already seamlessly integrated. If the loan is referred, UPST gets all three fees. If the loan is direct, then it doesn’t receive the referral fee. + 75% of loans are referred. Contracts with banks are typically twelve months in length that automatically renew. Contracts usually have minimum fee amounts but no limits/obligation on originations. Typically, after the loan is originated, ~20% is kept by the originating bank and ~80% is sold to institutional investors. The bank keeps it for a few days then sells to Upstart and, on the same day, Upstart sells to investors. Bank partners may also take an origination fee (~5%) from the amount borrowed which is deducted from the borrower’s principal.

UPL is a subset of consumer loans, which typically constitute a small part of a banks’ balance sheet. There’s ~$4.2tn in consumer credit outstanding and ~$150bn in UPL. The company serviced 300,000 loans in 2020, which translates into $3.6bn in loan originations – average loan size is ~$11,600. The average borrower is 28 years old, with a FICO score of ~690, ~20% APR and a fixed-rate ~5 year term. >50% of loans are made to people with at least a college degree. Most UPL are for refinancing credit card debt and debt consolidation. UPST’s average bank has ~$14bn in assets and does business in the North/Center of the country. There are 108 US banks with assets between $1-$100bn. In this sense UPST has ~5% share of those 108 banks. However, the total assets of UPST’s bank partners versus the total assets of all commercial banks in the US is ~1%.

Today UPST approves 70% of loans with no document upload or phone call, up from zero five years ago. It uses credit history and is checking information with other providers. Most fraud can be detected by asking borrower’s questions (and determining the truthfulness of the answers) to which UPST already knows the answer (borrower location, which browser he/she is using, etc.).

Strengths & Weaknesses

(1) Limited competition: I found certain evidence that indicates competition with UPST has been limited. UPST’s peers are different in the sense that some: (1) originate, fund the loan, and then sell it, (2) originate loans but are then obligated to buy them from the bank. No company is integrating their proprietary system into the banks like UPST. Furthermore, UPST’s “Risk Factors” section this section also makes several references to “potential competitors” as opposed to current competitors. These claims are further back by Paul Gu’s recent comments: “We see so little, so little happening in this direction, outside of what we’re building at Upstart”. It’s also not clear that other businesses are using as many variables and AI to the same extent UPST is.

(2) Distribution & 4,500 US Commercial Banks: In credit origination and in banking brand is of the uttermost importance. Partnering with banks allows for a sticky distribution. In addition to that it also avoids most balance sheet risk and reduces amount of capital in the business. Until a few decades ago US commercial banks could only operate in their own state. The result is that the US has many small commercial banks. The average commercial bank has ~$4.5bn. The mean, however, is not a good measure as assets are heavily skewed to the big banks, 78% of US commercial banks have <$1bn in assets. Naturally, these banks do not have the capital necessary to build out infrastructure like UPST did, particularly for small subset of their balance sheet like UPL. (3) Bank Partners Onboarding Process: The time between approaching a bank and booking revenue can take a long time. The reason for the long sales process is mostly related to: (1) skepticism about AI and, (2) getting approved through risk/credit committees. Technically implementing the software/hardware is not the reason. UPST targets (smaller) community and regional banks that tend to be much more worried about compliance than innovation.

More recently, UPST has launched an API for banks to integrate UPST’s credit model into their own systems in a simpler way. This seems to have helped in speeding up the process. Upstart started with Cross River Bank and has since added banks across the North and Center of the US. From 2015 to the end of 2020, it added twelve banks, and since then it has added another six. Over the last few years, the onboarding process has shortened as potential bank partners see peers’ results and have more confidence. Regional and community banks are having trouble with loan demand and that has also helped UPST start talking with them.

  1. Two customers account for +80% of revenue. Very dependent on two customers to fund loan volume.
  2. +50% of loans came from Credit Karma, which gives it bargaining power.
  3. Credit risk as they may be liable to repurchase loans. Repurchased ~0.3% of loans originated in 2020.
  4. Institutional investors not willing to buy securitized loans from Upstart.

The company made the first AI-trained loan in 2015, and this seems like an area which is ripe for change. Most US banks have a mobile app powered by third-party services, and Upstart is no different in the sense it’s also a third-party provider of a specific solution, and it only has eighteen banks. There’s a very long runway.

More recently the company has moved into auto loans (US commercial banks hold +$450bn in auto loans). The Prodigy acquisition serves to help them on this journey. The first step will be to get dealerships on Prodigy, the second step will be to try and cross-sell their credit software to dealerships.

However, it is also true that UPL is a cyclical industry. Most loan volume is for the purpose of consolidating/refinancing debt. The company has not operated through a period where the macroeconomic backdrop was persistently unfavorable. Although management seems perfectly aware of the inherent cyclicality and riskiness of UPL, I believe it is best to see how Upstart’s model works through more dire circumstances.

Furthermore, it is yet to be seen to what extent the ‘data advantage’ is actually a competitive advantage.