Upstart's Q3 2025: Contradictions Emerge on Model Conservatism, Auto Loan Growth, and Umi Adjustments

Date of Call: None provided

Financials Results

• Revenue: $277M, up 71% YOY and 8% sequentially
• EPS: $0.23 GAAP diluted EPS; $0.52 adjusted EPS (diluted)

Guidance:

• Q4 2025 revenue ~ $288M (fees ~$262M; net interest income ~$26M); contribution margin ~53%; GAAP net income ~ $17M; adjusted net income ~ $52M; adjusted EBITDA ~ $63M.
• Full year 2025 revenue ~ $1.035B (fees ~$946M; NII ~$89M); adjusted EBITDA margin ~22%; FY GAAP net income ~ $50M.
• Expect R&D balance reduction to accelerate in Q4 and into 2026; NII may moderate as balance sheet run-off occurs.
• Plan to modestly lower take rates to increase origination volume and customer LTV while maintaining fixed-cost discipline.

Business Commentary:

• Strong Application Volume Growth:
• Upstart reported a 30% increase in applications submitted in Q3, reaching over two million applications, the highest level in more than three years.

• The growth was driven by marketing programs and cross-selling efforts, despite a more conservative model that reduced conversion rates.

• Model Conservatism and Credit Performance:

• The company's model responded to macroeconomic signals by moderately reducing approvals and increasing interest rates, which led to a reduction in conversion rates from 23.9% in Q2 to 20.6% in Q3.

• The model's conservative behavior is attributed to an observed increase in the Upstart Macro Index (UMI) and not material deterioration in consumer credit strength.

• New Product Growth and Expansion:

• Upstart's newer products, including small-dollar loans, auto, and home, accounted for 12% of originations and 22% of new borrowers in Q3, with each product growing more than 300% year-on-year.

• This growth is attributed to improvements in products, partnerships, and expansion into new markets, notably in auto retail with more than doubled live lending rooftops.

• Profitability and Capitalization:

• Upstart reported a sixfold increase in GAAP net income for Q3 compared to the prior quarter, achieving profitability.
• This was driven by strong return performance on loans held directly on the balance sheet, despite a decline in conversion rates due to the model's conservatism.

  Sentiment Analysis:

  Overall Tone: Positive

  • Management highlighted 80% YOY transaction growth and 71% revenue growth, Q3 GAAP net income up ~6x sequentially, strongest application level in >3 years, and repeated statements of optimism for 2026 ("dramatically stronger company", "all systems go to finish the year strong").

Q&A:

• Question from Dan Doleav (Mizuho): Application demand looked very strong Q3 (apps +30% Q/Q) but guidance was below expectations—how do you reconcile strong applications with lower-than-expected originations/guidance?
  Response: Applications were strong but the models reacted to transient macro signals by tightening approvals and raising rates, lowering conversion; management says the conservatism has since abated and apps validate demand.

• Question from Kyle Peterson (Needham): Have recent negative auto headlines impacted your expansion plans or conversations with customers? And was the model tightness concentrated in the superprime (720+) segment?
  Response: No material impact—Upstart hasn't seen the fraud issues and has built processes to underwrite dealers; model conservatism affected multiple segments (not just superprime) and competition also plays a role in mid-700s.

• Question from Pete Christiansen (City): With improved marketing and more applications, has lead quality changed? Also, would an improvement in non-prime auto delinquencies materially drive Upstart auto originations?
  Response: AI-driven acquisition increased incremental applications and originations potential, but Q3 credit conservatism lowered conversion; auto credit performance is strong and market disruption could create opportunity—management is optimistic for auto growth in 2026.

• Question from Simon Clinch (Rothschild/Redburn): You indicate UMI improvement but guidance assumes some continued conservatism—will model conservatism persist into Q4? And is refinancing still the dominant driver of personal loan demand or is demand broadening?
  Response: UMI is improving but management will prudently allow lag before fully easing models so some Q4 impact persists; refinancing remains the dominant use case though personal loan use-cases are broadening.

• Question from Patrick Moley (Piper Sandler): How are talks with funding partners for R&D products trending and have private credit partners reduced demand given auto-sector noise?
  Response: Discussions are progressing well with strong partner appetite; diligence timelines lengthened (especially for auto) but no evidence of partner pullback—expect tangible agreements soon.

• Question from Mahir Bhatia (Bank of America): Was the conversion decline driven primarily by model conservatism? Paul, you mentioned reducing volatility—can you expand and is there a target conversion level? Also, why have repayment speeds increased and what are the credit implications?
  Response: Yes—model conservatism (fewer approvals, slightly higher rates, smaller sizes) was the dominant driver; calibration improvements reduce unwanted month-to-month volatility (~50% reduction in measurement error); faster repayment speeds appear broad (not just refi), likely reflecting improving consumer health but cause near-term pricing conservatism.

• Question from Reggie Smith (J.P. Morgan): Which component (approvals vs consumer acceptance) drove conversion changes? Are your offers shown alongside competitors when consumers shop? And how should we think about HELOC day-one take rates versus core personal loans? Any credit issues to flag?
  Response: Conversion changes were mainly approval-driven (model conservatism); offers can compete in various shopping contexts but mix varies by channel; HELOC take rates are expected materially lower (roughly ~half) than personal loans but on much larger average balances; credit performance remains exceptional.

• Question from James Fawcett (Morgan Stanley): Which specific model elements appeared conservative and how did you improve them? And how should we think about exit rates through Q4 into 2026?
  Response: Model separation remains best ever; the conservatism was a calibration response to short-term macro signals and sampling noise—work this quarter reduced sampling-driven variance; management expects solid Q4 and is optimistic on stronger 2026 growth as models and automation drive conversion improvements.

• Question from Rob Wildhack (Autonomous Research): Why does UMI suggest subprime is healthier while middle-prime shows elevated risk (the K-shaped observation)? And what drove sequential declines in Engineering and G&A spend?
  Response: Segmentation matters—sub-660 borrowers show relatively low UMI vs mid-prime (low-to-mid 700s) where defaults rose vs pre-COVID; hence a U-shaped pattern; lower OpEx driven by fixed-cost discipline and mechanical reductions in comp accruals tied to outlook adjustments.

• Question from John Hect (Jefferies): For HELOCs (avg ~$55–60k) what's the primary use case? Also, can you explain what your model is detecting that makes it differ from market signals (i.e., what is the 'black box' seeing)?
  Response: HELOCs are general-purpose (home improvement, debt consolidation, etc.); Upstart's models are built to detect cohort- and variable-controlled, real-time shifts (separation and calibration across thousands of features), enabling earlier, more precise signals than traditional backward-looking credit metrics.

Contradiction Point 1

Model Conservatism and Consumer Demand

It involves contradictory statements regarding the model's response to macroeconomic signals and consumer demand, which directly impacts the volume of transactions and investors' expectations.

How do you reconcile the strong application demand with the lower-than-expected transaction volume and guidance? - [Dan Dolev](Mizuho)

2025Q3: Applications grew about 30% quarter on quarter, but our model exhibited conservatism in response to macroeconomic signals, reducing approvals and raising interest rates. - [Dave Girouard](CEO)

Can you discuss conversion rate trends for approval and acceptance rates, and any changes due to new models? - [Reginald Lawrence Smith](JPMorgan Chase & Co)

2025Q2: The model advancements are detecting subtle credit relationships, leading to better approvals without a single dominant borrower characteristic change. - [Paul Gu](CTO)

Contradiction Point 2

Auto Loan Segment Growth

It involves differing perspectives on the growth and performance of the auto loan segment, which is a significant part of the company's business.

Have recent auto bankruptcies and credit issues affected your expansion plans? - [Kyle Peterson](Needham)

2025Q3: Upstart hasn't been directly impacted by these issues, but there's a broader caution in the market, and banks are exercising more diligence. - [Dave Girouard](CEO)

Why is the average loan size decreasing for personal loans, and what is the economic performance of new products like HELOC and auto loans? - [Kyle David Peterson](Needham & Company, LLC)

2025Q2: We continue to be really excited about the auto vertical and the opportunity it represents for us in terms of being able to expand our market opportunity, particularly into a much more consumer-centric experience. - [David J. Girouard](CEO)

Contradiction Point 3

Model Conservatism and Approval Rates

It highlights changes in the company's model behavior and approval rates, which directly impact revenue and investor expectations.

How do you reconcile strong application demand with lower-than-expected transaction volumes and guidance? - [Dan Doleav](Mizuho)

2025Q3: Applications grew about 30% quarter on quarter, but our model exhibited conservatism in response to macroeconomic signals, reducing approvals and raising interest rates. - [Dave Girouard](CEO)

Can you outline expectations for conversion rates and explain changes in the UMI slide? - [John Coffey](Barclays)

2025Q1: Conversion rates have grown significantly and are expected to be driven higher through model improvements and faster automated processes. - [Dave Girouard](CEO)

Contradiction Point 4

UMI and Model Behavior

It involves the company's response to adjustments in the Upstart Macro Index (UMI) and how it affects model behavior, impacting revenue and investor expectations.

Will the model's conservatism continue into Q4 despite UMI improvements? - [Simon Clinch](Rothschild and Company, Redburn)

2025Q3: The UMI improvements are materializing, but there's a lag in model adjustments. Some Q4 was impacted by the UMI rise, even though it's now subsiding. - [Sanjay Datta](CFO)

Can you discuss conversion rate expectations and address UMI slide changes? - [John Coffey](Barclays)

2025Q1: The Upstart Macro Index slide was revised to focus more on our data without directly referencing public Fed data. These changes don't impact our outlook, and we continue to use Fed data as an external reference. - [Sanjay Datta](CFO)

Contradiction Point 5

Model Conservatism and Borrower Demand

It involves contradictory statements regarding the model's response to macroeconomic signals and borrower demand, which directly impacts the volume of transactions and investors' expectations.

How do you reconcile strong application demand with lower-than-expected transaction volume and guidance? - [Dan Dolev](Mizuho)

2025Q3: Applications grew about 30% quarter on quarter, but our model exhibited conservatism in response to macroeconomic signals, reducing approvals and raising interest rates. - [Dave Girouard](CEO)

What caused the increase in borrower demand during Q4? - [Simon Clinch](Redburn Atlantic)

2024Q4: The increase in approvability and conversion is due to improved model accuracy, moderation in default rates, and the reflection of rate cuts from last fall in platform pricing. - [Sanjay Datta](CFO)