NASDAQ: UPST

Beyond core underwriting, we apply our proprietary AI models to other areas of our business, such as income and identity verification, fraud detection, and identifying loan stacking behavior, among others. The result is an exceptional digital-first experience with significant levels of automation that helps consumers throughout the loan process with minimal effort. In 2025, 91%1 of loans on our platform were fully automated, with no human intervention by Upstart. Consumer acquisition and targeting is another area where we apply our AI, making these activities increasingly efficient. Consumers primarily access Upstart-powered loans through Upstart.com and, for automotive retail in particular, through auto dealerships that use Upstart’s Auto Finance software.

Our AI models have been continuously upgraded, trained and refined for more than ten years. We believe the flywheel effects generated by our constantly improving AI models provide a significant competitive advantage.

Our dynamic marketplace allows us to serve borrowers across the credit spectrum. Loans issued through our marketplace are purchased by our network of institutional investors, retained or purchased by our lending partners, or in certain instances, held on our balance sheet. During the year ended December 31, 2025, out of the total principal of loan originations facilitated on our marketplace, 64% were purchased by institutional investors, 26% were retained or purchased by our lending partners, and 10% were held on our balance sheet. Investors may also invest in securities collateralized by Upstart-powered loans through our pass-through and securitization programs.

Institutional investors play an important role in our lending marketplace by providing capital for higher risk loans that may not be economically feasible for traditional banks and credit unions to hold. Today, more than 50% of the loan funding on our platform is through committed capital and other co-investment arrangements with institutional investors and lending partners, which provide valuable stability and resilience on the funding side of our platform.

1 See the section titled “Management’s Discussion and Analysis of Financial Condition and Results of Operations—Key Operating Metrics” for a definition of Percentage of Loans Fully Automated.

7

Table of Contents

We will continue working to secure additional capital to support the growth of our business and further diversify our institutional investor base and capital supply to ensure long-term scalability.

Our revenue consists primarily of fees paid by lending partners and institutional investors. We charge our lending partners platform and referral fees and the agreements with our lending partners may contain minimum fee amounts. In the year ended December 31, 2025, our top three lending partners collectively originated 83% of the loans facilitated through our marketplace and revenue from fees received from these lending partners accounted for 61% of our total revenue. We view the roles of these entities as largely interchangeable and value the resilience enabled by having a network of available lending partners. We also earn loan servicing revenue from contracts with lending partners and institutional investors.

As a usage-based platform, we target positive unit economics on each transaction, leading to a cash efficient business model with high margins. We believe these are the key components to achieve both high growth rates and profitability over time.

Our AI Lending Models

Our AI models are central to our value proposition for consumers and capital partners and are an important source of competitive differentiation. The key aspects of our AI models include:

Variables and Training Data

The volume of data utilized by Upstart’s AI models has grown significantly over time. Our personal loan underwriting model expanded from 23 variables at the end of 2014 to over 2,5002 at the end of 2025. Variables relate to a range of attributes, including credit experience, employment, educational history, bank account transactions, cost of living and loan application interactions. As of December 31, 2025, this model was trained on nearly 104 million repayment events, and it is only one of approximately 30 proprietary machine learning models that we developed in-house, which we use across our business.

The number of variables and volume of training data used in our models are co-dependent. The use of hundreds or thousands of variables is impractical without sophisticated machine learning algorithms to tease out the interactions among them, and sophisticated machine learning depends on large volumes of training data. Over time, we have been able to deploy increasingly sophisticated modeling techniques, leading to a more accurate system. This co-dependency presents a challenge to others who may aim to short-circuit the development of competitive models. While incumbent lenders may have vast quantities of historical repayment data, their training data lacks the hundreds of non-traditional variables that power our models.

Modeling Techniques

Growth in our training data has enabled the development of increasingly sophisticated modeling techniques. For example, while earlier versions of our models were centered on logistic regression and Monte Carlo simulations, our more recent models incorporate neural networks, gradient boosting, Markov Chain modeling, and Bayesian hyperparameter optimization. We expect that our data science investments and continued growth of training data will unlock even more powerful techniques over time.

2 “Variables,” often also called “features,” refers to raw variables and combined variables considered in our AI models. A “raw” variable is a non-combined, conceptually distinct unit of data, such as “applicant-reported savings.” A “combined” variable is data that has been transformed, combined, or otherwise engineered from a raw variable or set of raw variables, such as “applicant-reported savings” divided by “loan amount.”

8

Table of Contents

Model Applications

While our first model focused on predicting the likelihood of loan default, we have since applied models across the entire process of credit origination. These models quantify and reduce risk in various ways, while also increasing automation and funnel conversion.

Currently active AI models within the Upstart platform include:

•Acquisition targeting—identifies consumers likely to qualify for and have need for a loan;

•Loan stacking—identifies consumers likely to take out multiple loans in a short period of time;

•Time-delimited prepayment prediction—quantifies the likelihood that a consumer will fully prepay a loan earlier than originally scheduled;

•Income fraud—quantifies the risk of potential misrepresentation of borrower income;

•Identity fraud—quantifies the risk that an applicant is misrepresenting their identity; and

•Time-delimited default prediction—quantifies the likelihood of default for each period of the loan term.

Despite their sophistication, our AI models are delivered to lending partners in the form of a simple cloud application that shields borrowers from the underlying complexity. Additionally, our platform allows lending partners to tailor lending applications based on their policies and business needs. Within the construct of each lender’s self-defined lending program, our platform enables the origination of conforming and compliant loans at a low per-loan cost.

Our Ecosystem

Our ecosystem includes consumers and auto dealers, along with banks, credit unions and institutional investors who purchase Upstart-powered loans directly or invest in securities issued by our pass-through and securitization programs. This broad ecosystem allows participants to access and benefit from our products in a variety of ways, which can lead to broader adoption of our AI lending solutions.

Consumers

On the consumer side, we have built a mobile app and a mobile-responsive website (together, our “platform”) to aggregate demand on Upstart.com, where consumers are presented with loan offers. Consumers can quickly and easily inquire about a rate, evaluate and choose a loan offer, provide necessary information for verification and review required disclosures before final acceptance of the loan. We have also made significant investments in Upstart Auto Finance, a dealer-facing software-as-a-service application that modernizes the auto sales process for both the consumer and the dealer. Similar to Upstart.com, we expect Upstart Auto Finance to become an important aggregator of consumer demand.

Consumers on our platform are generally offered unsecured personal loans, secured auto loans, and HELOCs. Personal loans typically range from $200 to $75,000 in size, at APRs up to 35.99%, with terms typically ranging from three months to five years. Auto loans range from $3,000 to $60,000 in size, at APRs up to 29.99%, with terms ranging from two to seven years. HELOCs range from $26,000 to $250,000, at APRs up to 18.0%, with terms of 10 or 15 years. All loans feature a monthly repayment schedule and no prepayment penalty.

9

Table of Contents

Value Proposition to Consumers

•Best rates—An internal study, conducted and published in 2025, compared our personal loan AI model to that of a traditional underwriting credit score-based model. The study demonstrated the ability of our AI model to expand credit access to borrowers. Results from the study showed that our AI model approves 43% more borrowers and yields 33% lower average APR for approved loans.

•Best process—When consumers apply for a loan through our platform, the application experience is streamlined into a single process and the loan offers provided are firm. In the year ended December 31, 2025, 91% of Upstart-powered loans were fully automated, compared to approximately 70% at the time of our initial public offering in December 2020. Automation improvements were due in large part to product, engineering and machine learning enhancements to eliminate previously manual processes. This includes increasing the accuracy of our verification and fraud detection models to reduce human involvement, and removing inefficient or unnecessary processes and procedures.

Lending Partners and Institutional Investors

On the loan funding side, we target a wide range of small, medium, and large lending partners with an appetite to invest in improved underwriting and digital originations. Because AI is a new and disruptive technology, and lending is a traditionally conservative industry, we have brought our technology to the market in a way that allows us to grow responsibly and improve our AI models, while allowing lenders to take a prudent approach to assessing and adopting our platform. As of December 31, 2025, we had more than 100 lending partners. Our lending partners retain loans that align with their business and risk objectives. Because lenders vary with respect to program objectives, risk tolerance and funding capacity, program parameters can vary significantly across different lenders. Lending partners have access to reporting and other tools to manage their platform usage and portfolio. We also perform fairness testing on our models to help satisfy lending partners’ regulatory obligations.

Our lending partners control their programs when originating loans through our platform and do not solely rely on our models. Each lending partner sets and approves its own underwriting policy that establishes certain “hard” requirements or criteria, which may include minimum credit scores, minimum and maximum loan amounts, and maximum debt to income ratio. Upstart applies the lenders’ hard criteria prior to engaging its underwriting models. The majority of credit denials on the platform are due to the lending partners’ hard criteria from their underwriting policies. Borrower applications that meet a lender’s hard credit criteria are then assessed by Upstart’s underwriting model for default and prepayment probabilities, after which a pricing engine takes into account the underwriting model output in addition to pricing requirements set by lenders, such as target return objectives and maximum allowable APR limits. This process allows our lending partners to leverage our technology within the scope of their existing underwriting policies.

Loan volumes exceeding lending partners’ funding capacity or risk tolerance can be sold through our marketplace to our network of institutional investors with a broader and more diverse appetite for risk. As a result, we can develop our business and our AI models faster than if we relied only on the funding capacity of our lending partners. The combination of lending partners and institutional investors provides our lending marketplace with competitive and diverse capital.

Our network of institutional investors includes investors that purchase whole loans originated via Upstart’s platform, as well as investors that purchase securities, such as pass-through certificates. We are typically retained by institutional investors to service the loans originated on our platform. In the case of whole loan purchasers, we typically enter into loan purchase agreements and loan servicing agreements with such purchasers. We have pass-through certificate programs sponsored by certain financial institutions under which institutional investors can purchase securities collateralized by Upstart-powered loans from an issuer trust.

10

Table of Contents

While there are minimal differences between whole loan sales and sales of pass-through certificates from Upstart’s perspective, both programs are offered to provide flexibility to institutional investors in our marketplace. Some institutional investors may prefer pass-through certificates, which may be more liquid and require less operational complexity, while other institutional investors may prefer whole loan purchases, which are generally more cost-effective. Whole loans purchased after origination may later be included in our asset-backed securitization transactions whereby interests in these Upstart-powered loans are sold to other institutional investors.

For our asset-backed securitizations, we engage with investment banks to structure transactions under which we and/or certain of the purchasers of whole loans or pass-through certificates sell pools of whole loans to a bankruptcy-remote special purpose entity. The special purpose entities, through one or more intermediate transfers and entities, create and sell tranched asset-backed notes and subordinated certificates, in each case, backed by the collective pools of Upstart-powered loans sold into the investment structure.

Value Proposition to Lending Partners and Institutional Investors

•Competitive digital lending experience—We provide banks and credit unions with a cost-effective way to compete more effectively with competitors who have greater technology budgets.

•Expanded borrower base—We refer borrowers that apply for loans through Upstart.com to our lending partners, helping them grow both loan volumes and number of borrowers.

•Upstart referral network—Once we aggregate consumer demand on our platform, we pass those borrowers to our lending partners.

•Flexible configurations—We built a configurable lending solution designed to meet the needs of our lending partners. Because our lending partners have complete authority and control over their lending programs, they predetermine many aspects of their loan offerings, including interest rate and loan size ranges, target returns for various risk profiles, minimum credit score, maximum debt-to-income ratio, fee structures, and disclosures.

•Servicing—While most lending partners and institutional investors choose to have us service their loans (through a branded servicing portal), each has the option of directly servicing loans itself. Our servicing platform manages all communication with borrowers, credit reporting agencies, and when necessary, collections agencies.

•Delivering returns—We focus on credit performance compared to the expectations set by us at the time of origination. An equal investment in all vintages of Upstart-powered core personal loans originated in the fourth quarter of 2023 through the third quarter of 2025 is currently expected to deliver annual returns in line with a blended target of approximately 11.3% after servicing fees.

•New product offerings—While we continue to iterate and improve on our core Upstart-powered products, we have launched new credit products over the last several years, for example, to serve the auto and home lending markets and to provide a small dollar personal loan offering. Personal loans are one of the fastest-growing segments of credit in the U.S. while home lending is the largest segment of consumer lending, followed by auto financing. Our platform helps lenders provide a product their borrowers want, rather than letting borrowers seek loans from competitors. We continue to invest in expansion of our product offerings.

•Access to capital markets—We have built a broad and deep network of institutional investors who provide capital through purchases of whole loans, pass-through certificates, and asset-backed securitizations. We have secured multiple committed capital and co-investment arrangements with institutional investors, which deliver a significant amount of capital to the Upstart marketplace. In these arrangements, we share some of the risk, including the upside, in loan performance relative to our expectations. See the section titled “Management’s Discussion and Analysis of Financial Condition and Results of Operations” for more information regarding these risk sharing arrangements. We have continued our work to expand our relationships with institutional investors to deliver greater and more diverse capital to our marketplace.

11

Table of Contents

•Continuous engagement with rating agencies—Upstart-powered personal loans are analyzed by credit rating agencies and are subject to significant and constant scrutiny from experts. Credit ratings are often publicly available, which help institutional investors and lending partners gain confidence in Upstart-powered loans.

•Insights into changes in the economy—In 2023, we introduced the Upstart Macro Index (“UMI”), which estimates the impact of the macroeconomy on credit performance for Upstart-powered unsecured personal loans. UMI helps our lending partners and institutional investors better understand and account for the effect that macroeconomic conditions have on the credit performance of these loans.

Our Technology Infrastructure

Our cloud-based software platform incorporates modern technologies and software development approaches to allow for rapid development of new features.

Cloud-Native Technologies

We run our technology platform as containerized services on the Amazon Web Services cloud. Our architecture is designed for high availability and horizontal scalability. Our primary development platforms are Ruby on Rails and Python, but our Kubernetes-based compute environment gives us the flexibility to run heterogeneous workloads with minimal operational overhead. We deploy new software regularly without platform downtime, allowing borrowers and lenders to immediately benefit from the latest updates to our platform.

Data Integrity and Security

Our information security program governs how we safeguard the confidentiality, integrity, and availability of our consumer and capital partner data. Our environment is continuously monitored with a suite of tools designed to detect and respond to security events in both internal and user-facing systems. We have a robust secure software development cycle and regularly engage with third parties to audit our security program and to perform regular penetration tests of our Web application and cloud environment.

Configurable Multi-Tenant Architecture

Our multi-tenant architecture enables multiple lending partners to use the same version of our application while securely segmenting their data. Though all tenants are using the same version of our platform, our software is designed to be highly configurable to meet the needs of our diverse lending partners, allowing customizations to everything from the applicant user interface to the core rules governing credit decisioning.

Machine Learning Platform

In order to support innovation in our models, we have developed proprietary technologies to enable our machine learning team to develop, train, test and deploy new model updates with minimal engineering support. Our backend systems are designed to flexibly integrate with multiple third-party data sources to feed these models and support real-time decisioning.

Robust Reporting and Integration Capabilities

Our reporting application programming interfaces (“APIs”) provide institutional investors and lending partners the ability to access data through a programmatic interface. Our integration capabilities with lending partners include an ability to pre-fill applicant information via API and provide loan details in real-time to facilitate a seamless process from application to origination. Our lending partner reporting portal provides our lending partners with a centralized console to view daily performance metrics for their lending programs, review and verify their credit policy and program configuration, and access operational reports and documents on demand.

12

Table of Contents

Consumer Marketing

Our growth and marketing approach is driven by the strength of our product and the interest rates we offer. Over time, in large part due to the rapid and continuing improvements to our models, our ability to offer lower rates than our competitors has improved and we have been able to extend new loan offers to applicants who were previously not eligible or were quoted a higher rate.

Our growth and marketing initiatives are primarily focused on bringing potential borrowers to Upstart.com, where they can learn if they qualify for a loan and the terms of the loan offer in only a few minutes. Our customer acquisition channels combine a mix of online and offline, as well as paid and unpaid, channels. While we constantly experiment to expand and optimize our acquisition strategies, our largest channels include:

•Direct mail—We apply our strengths in data science to target individuals who both qualify for and may have a need for an Upstart-powered loan. The ability to analyze an individual’s credit data to target and mail prescreened offers of credit gives this channel a meaningful data advantage over other channels.

•Email marketing—We have an automated email program that sends customized messages and reminders to potential borrowers once they have created accounts to encourage them to complete their loan application.

•Organic traffic—As our brand recognition and reputation grow, an increasing number of potential borrowers come directly to Upstart.com. This is particularly the case for repeat borrowers.

•Online advertising—Search engines and social channels enable targeted outreach to potential borrowers with specific messages. In addition, we also advertise on streaming television services.

•Marketing affiliates—A variety of online media partners, such as loan aggregators, send us traffic on a cost-per-origination basis. Many loan aggregators also incorporate application data to provide online prescreened offers, which leads to highly targeted and interested referrals. For example, a significant number of consumers that apply for and obtain a loan on Upstart.com learn about and access Upstart.com through the website of one of our partners, Credit Karma.

Operations

We have developed sophisticated tools that our internal operations team uses to support the origination and servicing of credit. Our operations teams, including credit analysts, fraud specialists, customer support, payments specialists, and supporting services (like quality assurance and training) work to deliver a seamless user experience to consumers on behalf of our capital partners.

Loan Origination Operations

While verification is primarily and increasingly handled by our software and AI models, we also offer Upstart-designed tools to guide credit analysts and fraud specialists in cases where our software is not yet able to sufficiently verify borrower information. By providing a prescriptive and unique path for each applicant, our system helps our operations team provide a streamlined experience for as many borrowers as possible.

This team focuses on the small minority of borrowers whose applications are not entirely automated or any applicant who has questions or issues throughout the application process, while expediting the approval process to the extent possible, and identifying and rejecting fraudulent applications. Our operations team works closely with our engineering and machine learning teams to further increase our levels of automation.

Most prospective borrowers and applicants interact with Upstart via our online platform and help center, but we also make agent-based support readily available to all borrowers. For phone support, we partner with external call center vendors and have a team of dedicated Upstart agents with specialized training.

13

Table of Contents

Servicing Operations

Upstart-powered loans, with the exception of HELOCs, are serviced via our proprietary platform. For borrowers who miss payments, we focus on early intervention and attempt to reach them via emails, calls, texts, and mail to help bring their account current or offer hardship options in accordance with the creditor’s servicing policies. We also use AI, including agentic AI, to help with customized servicing interventions and improved servicing efficiency throughout the loan lifecycle. Borrowers on our platform are supported via a combination of internal and third-party agents.

We hold collections licenses in the majority of states and conduct first-party collections activities. We also partner with third-party agencies for collections, including for loans that are charged off. Debt collection calls and collection performance are reviewed regularly by our quality assurance or vendor management teams. We also perform collections agency onsite audits annually.

Competition

Consumer lending is a vast and competitive market, and we compete in varying degrees with all other sources of unsecured and secured consumer credit, including banks, non-bank lenders (including retail-based lenders), and other financial technology lending platforms. Because personal loans often serve as a replacement for credit cards, we also compete with the convenience and ubiquity that credit cards represent.

On the lending partnership side, we compete with a variety of technology companies that aim to help lenders with the digital transformation of their business, particularly with respect to all-digital lending. This includes new products from legacy lending technology providers as well as newer companies focused entirely on lending software infrastructure for lenders. We may also face competition from lenders or companies that have not previously competed in the consumer lending market, including companies with large and experienced machine learning teams and access to vast amounts of consumer-related information that could be used in the development of their own credit risk models.

We believe we compete favorably based on the following competitive factors:

•Constantly improving AI models;

•Compelling loan offers to consumers that improve regularly;

•Automated and user-friendly loan application process;

•Cloud-native, multi-tenant architecture;

•Combination of technology and customer acquisition for lending partners;

•Robust and diverse lending marketplace; and

•Brand recognition and trust.

Government Regulation

We and the loans made through our platform by our lending partners are subject to extensive and complex rules and regulations and examination by various federal, state and local government authorities. While compliance with such requirements is at times complicated by our novel business model, we have resources and processes to help us comply with these rules and regulations.

We are currently, and expect in the future, to be subject to laws and regulations administered by the Consumer Financial Protection Bureau (the “CFPB”). In addition, the Federal Trade Commission (“FTC”) has jurisdiction to investigate aspects of our business. Other state and federal agencies, including prudential bank regulators, state regulatory bodies, and state attorneys general have the ability to regulate aspects of our business

14

Table of Contents

directly or through our lending partners. Further, we are subject to inspections, examinations, supervision and regulation by applicable agencies in each state in which we are licensed to originate (in the case of HELOC), broker, purchase, and/or service loans. Regulatory oversight of our business may change over time.

Below, we summarize several of the material federal and state lending, servicing and consumer protection related laws applicable to our business. For more information regarding the various federal and state laws and regulations to which we are subject or may become subject, see