1. Investment Snapshot
2. Thesis
3. Valuation & Price Target
4. Business & Product Moat
5. People & Governance
6. Market & Macro
7. Financial Quality
8. Risk Register
9. Prediction Market
10. 𝕏 Posts
Discussion
1. Investment Snapshot
2. Thesis
3. Valuation & Price Target
4. Business & Product Moat
5. People & Governance
6. Market & Macro
7. Financial Quality
8. Risk Register
9. Prediction Market
10. 𝕏 Posts
Discussion
1. Investment Snapshot
2. Valuation
Discussion
Symbol
UPST
Event Date
2020-12-16
Sector
Financials
Subsector
Financial Services
Offer Range
$20.00-$20.00
Shares Offered
12.02M
76.9M
$1.5B
15.6%
Implied Upside vs Midpoint
Description
Our mission is to enable effortless credit based on true risk. We are a leading, cloud-based artificial intelligence lending platform. Artificial intelligence, or AI, lending enables a superior loan product with improved economics that can be shared between consumers and lenders. Our platform aggregates consumer demand for high-quality loans and connects it to our network of Upstart AI-enabled bank partners. Consumers on our platform benefit from higher approval rates, lower interest rates, and a highly automated, efficient, all-digital experience. Our bank partners benefit from access to new customers, lower fraud and loss rates, and increased automation throughout the lending process. Since inception, our bank partners have originated over 620,000 personal loans that have generated more than 9 million repayment events. In the nine months ended September 30, 2020, approximately 70% of Upstart-powered loans were entirely automated. Credit is a cornerstone of the U.S. economy, and access to affordable credit is central to unlocking upward mobility and opportunity. The FICO score was invented in 1989 and remains the standard for determining who is approved for credit and at what interest rate.(1) While FICO is rarely the only input in a lending decision, most banks use simple, rules-based systems that consider only a limited number of variables. Unfortunately, because legacy credit systems fail to properly identify and quantify risk, millions of creditworthy individuals are left out of the system, and millions more pay too much to borrow money.(2) The first generation of online lenders focused on bringing credit online. Analogous to earlier internet pioneers, these companies made shopping for and accessing credit simpler and easier for consumers and businesses. It was no longer necessary to stand in line at a bank branch, to sit across the desk from a loan officer and to wait weeks or months for a decision. These lenders enabled the emergence of personal loan products that were previously unprofitable for banks to offer. While they brought the credit process online, they inherited the decision frameworks that banks had used for decades and did not address the more rewarding and challenging opportunity of reinventing the credit decision. We leverage the power of AI to more accurately quantify the true risk of a loan. Our AI models have been continuously upgraded, trained and refined for more than eight years. We have discrete AI models that target fee optimization, income fraud, acquisition targeting, loan stacking, prepayment prediction, identity fraud and time-delimited default prediction. Our models incorporate more than 1,600 variables and benefit from a rapidly growing training dataset that currently contains more than 9 million repayment events.(3) The network effects generated by our constantly improving AI models provide a significant competitive advantage—more training data leads to higher approval rates and lower interest rates at the same loss rate. We have been able to demonstrate through several studies that AI lending works. First, in 2019 the Consumer Financial Protection Bureau, or CFPB, reported that a study by Upstart of its data using a methodology specified by the CFPB showed that our AI model approves 27% more borrowers than a high-quality traditional model, with a 16% lower average APR for approved loans.(4) Second, when compared to credit models from several large banks, our AI models approve approximately 2.7 times as many borrowers at the same loss rate.(5) Third, for pools of securitized loans, our realized loss rates were only approximately half of those predicted by Kroll, a prominent credit rating agency; over that same time period, realized losses for the same pool of loans were on average only 5% different than our internal forecasts.(6) And finally, we regularly monitor the accuracy of our AI models in comparison with simple credit score-based models and have observed higher model accuracy across a variety of statistical measures relating to each model’s predictive accuracy.(7) Our AI models are provided to bank partners within a consumer-facing cloud application that streamlines the end-to-end process of originating and servicing a loan. We have built a configurable, multi-tenant cloud application designed to integrate seamlessly into a bank’s existing technology systems. Our highly configurable platform allows each bank to define its own credit policy and determine the significant parameters of its lending program. Our AI models use and analyze data from all of our bank partners. As a result, these models are trained by every Upstart-powered loan, and each bank partner benefits from participating in a shared AI lending platform. Consumers can discover Upstart-powered loans in one of two ways: either via Upstart.com or through a white-labeled product on our bank partners’ own websites. Loans issued through our platform can be retained by our originating bank partners, distributed to our broad base of approximately 100 institutional investors and buyers that invest in Upstart-powered loans or funded by Upstart’s balance sheet. In the third quarter of 2020, 22% of the loans funded through our platform were retained by the originating bank and 76% of loans were purchased by institutional investors through our loan funding programs. Our institutional investors and buyers that participate in our loan funding programs, which include Goldman Sachs, PIMCO and funds managed by Morgan Stanley Investment Management, invest in Upstart-powered loans through whole loan purchases, purchases of pass-through certificates and investments in asset-backed securitizations. We enter into nonexclusive agreements with our whole loan purchasers and each of the grantor trust entities in our asset-backed securitizations, or ABS, under which our ABS investors benefit from our loan servicing capabilities. The remaining 2% of loans funded through our platform in the third quarter of 2020 were funded through our balance sheet. Our revenue is primarily comprised of fees paid by banks. We charge banks referral fees for each loan referred through Upstart.com and originated by a bank partner, platform fees for each loan originated (regardless of its source) and loan servicing fees as consumers repay their loans. Our agreements with our bank partners are nonexclusive, generally have 12-month terms that automatically renew, subject to certain early termination provisions and minimum fee amounts, and do not include any minimum origination obligation or origination limits. As a usage-based platform, we target positive unit economics on each transaction, resulting in a cash efficient business model that features both high growth rates and profitability. As of September 30, 2020, we had 10 bank partners. In the nine months ended September 30, 2020, Cross River Bank originated 72% of the loans facilitated on our platform and fees received from Cross River Bank accounted for 65% of our total revenue. Our current agreement with Cross River Bank began on January 1, 2019 and has an initial four year term, with a renewal term for an additional two years following the initial four year term. (1) Rob Kaufman, myFico Blog: The History of the FICO Score, August 2018. (2) Patrice Ficklin and Paul Watkins, Consumer Financial Protection Bureau Blog: An Update on Credit Access and the Bureau’s First No-Action Letter, August 2019. (3) References to variables in this prospectus refer to all raw variables and certain combined variables considered in our AI models. (4) Ficklin and Watkins; (5) In an internal study, Upstart replicated three bank models using their respective underwriting policies and evaluated their hypothetical loss rates and approval rates using Upstart’s applicant base in late 2017. Such result represents the average rate of improvement exhibited by Upstart’s platform against each of the three respective bank models. (6) In an internal study, Upstart compared the actual realized loss rates of Upstart loans securitized in five securitization transactions between June 2017 and September 2019 and the loss rate predictions for those loans obtained from KBRA Surveillance Reports published by Kroll Bond Rating Agency in December 2019. As compared to Kroll’s loss predictions, actual realized losses were approximately 31% to 71% lower, with an average deviation across all five securitization transactions of -48%. As compared to our internal forecasts, actual realized losses ranged from approximately 35% higher (for the earliest securitization transaction) to approximately 17% lower (for the most recent securitization transaction), with an average absolute deviation across all of five securitization transactions of approximately 13%. (7) Upstart compares on a monthly basis its AI models to (i) a FICO-only model and (ii) a “FICO+” model, which considers loan amount, debt-to-income ratio, monthly income, number of inquiries and number of trade accounts in addition to FICO score, which we believe is representative of the model many of our sophisticated competitors would use. To conduct a comparison of the Upstart AI models to the models described in (i) and (ii), we run applicant information through the Upstart AI models, the FICO-only model, and FICO+ model, comparing performance by analyzing five commonly-used statistical metrics, each of which measures the deviation between predicted losses and actual losses for each model. These metrics include: the root mean square error of net present value of losses, normalized logistic loss, ratio of the gini coefficients of predicted to observed rankings, the area under the receiver operating characteristics curve, and the Kolmogorov-Smirnov statistic. --- Upstart Network, Inc. was incorporated in Delaware in 2012. Pursuant to a restructuring, Upstart Holdings, Inc. was incorporated in December 2013 and became the holding company of Upstart Network, Inc. Our principal executive offices are located at 2950 S. Delaware Street, Suite 300, San Mateo, California 94403, and our telephone number is (650) 204-1000.Our website address is www.Upstart.com.
Upstart Holdings, Inc. annual income statement and balance sheet, FY 2021 to FY 2025, as reported in SEC filings.
| Metric | FY 2021 | FY 2023 | FY 2024 | FY 2025 |
|---|---|---|---|---|
| Revenue | $849M | $514M | $637M | $1.0B |
| Operating income | $141M | ($257M) | ($173M) | $42.6M |
| Net income | $135M | ($240M) | ($129M) | $53.6M |
| Metric | FY 2021 | FY 2023 | FY 2024 | FY 2025 |
|---|---|---|---|---|
| Total assets | $1.8B | — | $2.4B | $3.0B |
| Total liabilities | $1.0B | — | $1.7B | $2.2B |
| Total equity | $807M | $635M | $633M | $799M |
| Cash & equivalents | $1.2B | $468M | $788M | $652M |