← Back to list
Colossus (Invest Like the Best / Business Breakdowns)Podcast30 Sep 2021Source: joincolossus.comHost: Patrick O'Shaughnessy

Dave Girouard - Making Better Loans - [Founder’s Field Guide, EP. 52]

In plain words

This is about Upstart founder Dave Girouard on using AI to make loans fairer. He says the old FICO credit score (a 3-digit number) is outdated; AI can use more data to predict risk better, helping more people get loans. He's bullish on Upstart, where 70% of loans are auto-approved in seconds. He warns Visa and MasterCard face threat from BNPL (buy now, pay later) like Afterpay, but Square's purchase of Afterpay is still uncertain.

AI SummaryAI-generated · may contain errors · verify against the original

Upstart co-founder and CEO Dave Girouard discussed on the program how the AI-driven lending platform improves loan accessibility and affordability through machine learning models. The core argument is that the lending market and its opportunity are underestimated and should not be viewed merely as a

~10 min full read · 9 sections
Deep Analysis

Dave Girouard - Making Better Loans - [Founder’s Field Guide, EP. 52]

At a Glance

Dave Girouard (Co-founder and CEO of Upstart) discusses in the program how AI-driven lending platforms use machine learning models to improve loan accessibility and affordability. The core argument is that the lending market and its opportunities are severely underestimated and should not be viewed merely as a customer acquisition tool — Girouard points out that the net interest income generated by lending in the U.S. is comparable to the profits of the entire technology industry, a scale far beyond most people's perception.


Theme 1: Speed as a Habit – A Three-Tier Acceleration Mechanism for Decision-Making

Girouard argues that the essence of corporate speed lies in accelerating the three-tier chain of "decision-making, execution, and persuasion," rather than simply pursuing speed for its own sake.

  • Decision Filtering: The CEO's goal is not to make more decisions, but to ensure that important decisions are made at lower levels of the organization. Girouard emphasizes: "If you are involved in every decision the company makes, that is a bad sign." The key is to distinguish which decisions are worth two minutes and which are worth two days, while avoiding the pursuit of consensus—"Look at all the statues in city parks; not a single one was erected by a committee."
  • Execution Acceleration: The core is to challenge assumptions about time. Girouard recalls that during his time at Google, Larry Page's classic question—whether the team said two months or six months—was always, "Why can't it be done in a week?" This forced the team to expose hidden serial assumptions and instead think about parallel execution.
  • External Persuasion: For partners/suppliers, Girouard's strategy is to "make it impossible for them to refuse from their standpoint"—using reason, logic, and goodwill to get the other party to voluntarily accelerate, rather than applying forceful pressure.

Girouard's Acknowledged Uncertainty: Speed is not a panacea; excessive pursuit of it can lead to wrong decisions, but "a timely decision is almost always better than a delayed correct one."


Theme 2: The Core of AI Lending — From FICO to 360-Degree Risk Profiling

Girouard argues that the FICO scoring system is 30 years outdated, and AI models can significantly improve credit prediction accuracy through more dimensional data, while simultaneously enhancing loan accessibility across all population groups.

  • Limitations of FICO: The three-digit score born in 1989 cannot capture a person's full credit profile. Girouard points out that a Federal Reserve study shows "30% of African Americans have FICO scores in the lowest decile," indicating a structural bias in the system.
  • Upstart's Model Mechanism: Expanding from "three digits" to "12–60 variables," including education level, field of study, industry, employer, etc. The core of the model is to predict monthly cash flow — "If you have a three-year loan, that's 36 months. The model needs to predict in which month you will default or prepay."
  • Synergy Between Data and Algorithms: Girouard explains that more complex algorithms require more training data — Upstart uses 40,000–50,000 monthly repayment or default data points as training material, and the model continuously iterates as data accumulates.
  • Counterintuitive Discovery: Early focus on young borrowers with "thin credit files" — Girouard's 20-year-old co-founder (Yale graduate, perfect SAT score, six-figure income, zero debt) was rejected by multiple institutions or offered only a $10,000 loan at 24% interest — this became Upstart's starting point.

Falsification Condition: If the model shows systematic bias in a certain demographic group, Upstart can revert to an older version — the company submits bias test data for all applicants to the CFPB every quarter.


Theme 3: Business Model — Not a Bank, but the AI Brain for Banks

Girouard explicitly stated that Upstart's decision to become a technology partner for banks rather than a bank itself defines the company's growth trajectory.

  • Revenue Structure: Approximately 97% of revenue comes from three types of fees paid by banks — referral fees, platform fees (for managing loan origination on behalf of banks), and loan servicing fees. Banks retain all the "knobs and dials" (which Girouard calls "geek knobs") for loan decisions, allowing them to independently set target return rates and risk preferences.
  • Value Proposition: Banks can choose between two extreme strategies — maintaining the approval rate while reducing losses by approximately 75%, or maintaining the loss rate while increasing the approval rate by 2-3 times. Girouard believes the "middle ground" is the norm.
  • Customer Experience: The median borrower is in their 20s, and approximately 70% of loans are "fully automated with no human intervention" — receiving a rate quote within seconds of application, instant approval, with no need to upload documents or make phone calls.

Girouard's Self-Defense Note: He acknowledges that "convincing the world that AI can be fairer" is the biggest challenge. Readers should note that this is a long-side perspective — regulatory and consumer protection group skepticism is real.


Theme 4: Structural Opportunities in the Lending Market—Two Orders of Magnitude in Inefficiency

Girouard argues that the inefficiency in the lending industry is far greater than most imagine—"90% of interest payments are completely unnecessary, stemming from sluggish risk models."

  • Magnitude Comparison: Girouard compares Upstart's opportunity to Renaissance Technologies' high-frequency trading—"They try to shave off 3 basis points, while we face two orders of magnitude of inefficiency."
  • Market Structure: Personal loans (approximately $100 billion annually) are the fastest-growing category but remain far smaller than auto loans (approximately $700 billion) and mortgages (an even larger magnitude). Girouard observes that unsecured loans are gaining market share from secured loans (such as home equity loans) and credit cards—"People would rather pay a bit more interest for a $10,000 loan that arrives in 10 days than wait two months for a home equity loan."
  • Threat from Buy Now Pay Later: Girouard believes that after Afterpay's acquisition by Square, BNPL players have gained a "two-sided network," potentially posing a real threat to Visa/MasterCard's payment rails—but the key question is "whether the fees merchants actually pay are higher than those for Visa/MasterCard," which determines the sustainability of this model.

Falsification Condition: If BNPL merchant fees remain persistently higher than those of traditional payment networks, or if a price war leads to margin compression, the disruptive potential of this model will diminish.


Theme 5: Founder CEO vs. Professional CEO — Differences in Decision-Making Authority

Girouard argues that founder CEOs possess an unparalleled advantage over professional CEOs in terms of decision-making speed and authority, particularly when a company needs to pivot quickly.

  • Specific Case: During COVID, Upstart completely changed its strategic direction within three months — "If I were a CEO hired by the board, they would think, 'What is this guy doing?' But as a founder, I can directly say, 'No, I know this company. It is me.'"
  • Comparative Perspective: Girouard cites DoorDash (founder-led) and Uber (professional CEO-led) as examples, arguing that the former has a "huge advantage" in decision-making flexibility.
  • Public Market Difference: Girouard finds that public market investors are more willing to accept Upstart than private market investors — "The private market never really understood us, or simply didn't like us." Public investors can "buy today and sell tomorrow," so they focus more on growth and profit data itself rather than the narrative.

Girouard's Self-Reflection: He acknowledges that Upstart's difficulty in raising funds in the private market may stem from itself — "We were too introverted and failed to tell a compelling story from an industry perspective."


Mentioned Positions

Position Guest Stance Key Data
Upstart Bullish (Founder’s perspective) ~97% of revenue from bank fees; 70% of loans processed fully automatically; median borrower age in their 20s; 40,000–50,000 repayment/default data points per month
Visa/MasterCard Risk warning BNPL may pose a threat, but "the strong network effect has existed for a long time and is almost unbreakable"
Square (acquired Afterpay) Neutral observation Has a two-sided network, may pressure traditional payment rails
DoorDash Positive assessment Founder-led, with decision-making advantages
Uber Neutral assessment Led by a professional CEO, with "dynamic differences"
Renaissance Technologies Comparative reference Attempts to shave off 3 basis points, while the lending industry has "two orders of magnitude of inefficiency"

Judgments Worth Remembering

1. "90% of interest payments are completely unnecessary" (Girouard) — stemming from the sluggishness and inefficiency of risk models, not the actual credit risk of borrowers.

2. "The net interest income from lending is comparable to the profits of the entire tech industry" (Girouard) — this magnitude comparison reveals the true scale of the lending market, far beyond what can be summarized as a "customer acquisition tool."

3. "A timely decision is almost always better than a delayed correct decision" (Girouard) — the core of the speed habit is "first decide how much time to spend on making a decision," rather than pursuing perfection.

4. "Larry Page would ask: Why can't it be done in a week?" (as recalled by Girouard) — regardless of whether the team reports 2 months or 6 months, this question exposes hidden serial assumptions and forces the team to consider parallel execution.

5. "30% of African Americans have FICO scores in the lowest decile" (Girouard citing Federal Reserve research) — the FICO system itself has structural biases, and AI models have the potential to become an "equalizing tool."

6. "We never received more than one valid lead term sheet" (Girouard) — Upstart faced extreme difficulty raising funds in the private market and ultimately proved itself through the public market — "Public investors can buy today and sell tomorrow, so they focus more on the data itself."

7. "A founder-CEO can completely change direction within three months, while a professional CEO would be questioned" (Girouard) — Upstart's strategic pivot during COVID is a classic case of founder authority.

8. "AI records the past, AI predicts the future — the two are naturally related" (Girouard) — the combination of immutable records from encryption and AI's predictive capabilities may give rise to better credit products, especially in cross-border scenarios.