Matt Clifford – Investing Pre-Company
- [Invest Like the Best, EP.154]
At a Glance
Matt Clifford is the co-founder of Entrepreneur First (EF), which pioneered the "talent investing" model—investing in founders before they have even formed a company. The core thesis of this episode: Talent is the most undervalued scarce resource globally, and the historical evolution of the "technology of ambition" determines who, when, and how gains the greatest leverage. The most impactful judgment in the entire episode: "The pricing power dilemma of traditional venture capital stems from its default path—founders must already have a team and an idea; EF moves the investment decision forward to 'Day Zero,' replacing two meetings with 100 days of dense data points to create a scalable judgment system."
Theme 1: The History of Technological Ambition—Talent Flows Shape Economic Structures
Matt Clifford argues that every era has a "technology of ambition"—the tool that best empowers individuals—and whoever masters it shapes the economic structure of that era.
- Historical Context: 1,000 years ago, literacy was the technology of ambition—Thomas Wolsey (son of a butcher) learned to read and write through church schools, eventually becoming the most powerful man in England (builder of Hampton Court Palace). 500 years later, military command became the lever—Napoleon rose to general through the Paris Military Academy. In the 20th century, finance became dominant—writing a check in New York could influence the world. Today, tech entrepreneurship offers unprecedented leverage—Napoleon would be "green with envy" at Mark Zuckerberg's reach.
- Mechanism Breakdown: Clifford notes that Silicon Valley is special not because its people are genetically different, but because "entrepreneurship" is the default answer for the smartest people there. Elsewhere, the answers differ: top graduates from Cambridge/Oxford in the UK flow into finance (Goldman Sachs), elites from the National University of Singapore enter government (Prime Minister's Office), and graduates from France's Grandes Écoles become civil servants. "If you multiply this question by a country's population, you get that country's industrial structure."
- Implication: EF's mission is to become the "missing institution"—akin to universities for literacy or military academies for command—making tech entrepreneurship the default path for the world's smartest talent. Falsification condition: If EF fails to consistently attract top talent (i.e., its "flywheel" reverses), the model will break down.
Theme 2: How to Identify "Zero-Day" Talent — 100 Data Points vs. Two Meetings
Clifford's core methodology: EF does not attempt to "match" founders, but instead creates a large number of teams and lets the weak ones naturally break apart; those that remain are the good ones.
- Mechanism breakdown: EF's process is structured as "3+3" months. In the first three months, individuals join with the goal of finding a co-founder. EF does not engage in "top-down matching" (a lesson learned from early failures), but instead encourages rapid team formation, rapid testing, and rapid "twisting." Key principle: "Don't overthink team formation; use productivity as the only metric. No good team is unproductive, and almost no productive team is bad." In the latter three months, the process resembles a more traditional accelerator, helping teams develop business models, find early customers, and prepare for seed rounds.
- Data chain: EF has helped 1,000 individuals found 200 companies, with a total valuation of $1.5 billion. Within EF, 40-50% of individuals receive investment after three months (i.e., EF exercises its options); of those, 60-70% secure seed rounds from institutional investors within six months. For comparison: in traditional seed funds, approximately 20% of companies lose a co-founder before their Series A; within EF, this figure is only 10%.
- Unique judgment: Clifford emphasizes that EF's investment decision is not "I have seen this team and done due diligence," but rather "Do I trust the data generating process and the stack rank?" He shared a counterintuitive case: the Acurix team (Jacob and Lawrence) conducted an intense series of doctor visits (approximately 100 meetings) in the first 100 days of EF, but their business model was unclear, and they raised only £275,000 (the average seed round in EF London is £1.4 million). Clifford privately "downgraded" the team, but the data (productivity ranking) was always correct. Within nine months, Acurix entered 30% of UK clinics, now covers 50%, has sent text messages to over 8% of the UK population, and completed a $10 million Series A round. Conclusion: "Every time I added my own 'signal' beyond the data, I was wrong."
Theme 3: Deep Tech — A Natural Ally for Talent Investing
Clifford argues that deep tech aligns far better with talent investing than consumer internet, as it inherently mitigates the adverse selection problem.
- Mechanism Breakdown: EF defines "deep tech" as companies where the answer to "why now" stems from technological change (rather than platform change, social change, or regulatory change). Three features make deep tech suitable for talent investing: 1) Clear problem definition (e.g., "Can we achieve 70% better video compression?" — with a well-defined market); 2) Team quality is the best predictor of outcomes (very few people globally can solve the problem); 3) Objective success criteria (speed, quality, or cost).
| Dimension |
Consumer Internet (e.g., Snap/Instagram) |
Deep Tech (e.g., Magic Pony Technology) |
| Initial assessment difficulty |
Extremely high — impossible to predict winners on day zero |
Relatively low — problem is clear, team quality is key |
| Capital requirements |
Low — MVP can be built quickly |
High — requires more time and capital |
| Adverse selection risk |
High — good CEOs don't need EF |
Low — good CEOs still need EF to find a CTO |
| Time horizon |
Fast (results in months) |
Slow (3 years with no revenue is normal) |
- Implication: Clifford believes that the next layer of "software is eating the world" is "machine learning is eating the world" — biology, chemistry, energy, manufacturing, healthcare, and other fields will generate trillions of dollars in value. EF's allocation: 70% software, 20% validated non-software technology, 10% fully asymmetric "wild cards" (e.g., novel silicon photonics companies). Falsification condition: If the commercialization cycle for deep tech companies far exceeds expectations (e.g., >5 years), EF's capital efficiency will be challenged.
Theme 4: Geographic Arbitrage — The "Absurd" Mispricing in Seed Round Valuations
Clifford presents a contrarian view: the impact of geography on seed round pricing is severely overestimated, creating significant arbitrage opportunities.
- Data Chain: The biggest driver of global seed round pricing is geography — a Singapore-based company's seed round valuation is roughly 50% of a London-based one. The conventional view attributes this to exit values also being 50% lower, but Clifford counters: "That's looking at data through the rearview mirror. Deep tech companies are global winners; they don't deserve to be worth half just because their headquarters are in Singapore."
- Mechanism Breakdown: EF discovered an overlooked talent network in Singapore: approximately 500 Iranians (concentrated in STEM graduate programs at the National University of Singapore), of which EF has funded about 10% (roughly 50 individuals). This stems from "dislocated networks" — Iranian students, restricted by visa policies from going to the U.S., find Singapore the best alternative. "We didn't set out to invest in Iranian graduate students; we spent a lot of time on campus, and they became a magnet."
- Implication: Clifford suggests that if he were to start a new VC, he would "launch a deep tech seed fund in Southeast Asia and write checks within a week" — speed being the quality founders value most (EF's internal data shows that when founders evaluate VCs, "speed" is the number one driver, surpassing hiring support, market access, etc.). Falsification condition: If Southeast Asian deep tech companies' exit values persistently lag global levels, geographic arbitrage will cease to exist.
Theme 5: AI Nationalism and the Geopolitical Leverage of Talent
Clifford argues that AI is the first strategic technology where "the private sector holds an absolute talent advantage," overturning traditional national security paradigms.
- Mechanism Breakdown: Ian Hogarth's "AI Nationalism" framework posits that AGI (Artificial General Intelligence) is not a "faster internet" but a technology that alters all domains of geopolitical competition. Analogy: If the Manhattan Project had been led by General Electric (rather than the U.S. government), we would say "it should be immediately nationalized." Today, DeepMind, OpenAI, and Google hold the world's top AI talent—the NSA monopolized cryptography talent 20 years ago, but today Google hires away most of the top machine learning PhDs.
- Data Chain: Clifford notes that talent is now the most effective constraint on big tech companies—Google and Facebook can withstand confrontation with the U.S. government, but they cannot withstand confrontation with 100 of their best machine learning engineers. Project Maven (Google's collaboration with the Department of Defense) was terminated due to employee protests, exemplifying this dynamic. "Imagine you are a Chinese intelligence officer tasked with ensuring China's long-term strategic advantage in AI—the smartest move is to instill in liberal Google engineers the belief that 'cooperating with the U.S. government is evil.' In China, this would not happen."
- Projection: Clifford believes that "technological sovereignty" will become a core issue over the next decade—nations must possess independent AI capabilities or risk becoming "AI client states." Falsification condition: If AI development does not reach AGI levels, or if the cooperation model between the private sector and governments undergoes a fundamental shift (e.g., large-scale government recruitment of AI talent), this judgment will weaken.
Mentioned Positions
| Position |
Analyst View |
Key Data |
| Magic Pony Technology |
Bullish (one of EF's best cases) |
Video compression technology, acquired (undisclosed amount) |
| Acurix |
Bullish (productivity ranking validation case) |
Raised £275k (seed round) → $10M (Series A); covers 50% of UK clinics, reaches 8% of population |
| Cloud NC |
Bullish (vertical integration model) |
End-to-end automated subtractive manufacturing, started from software then built its own factory |
| Tractable |
Bullish (deep tech case) |
Computer vision for insurance, zero revenue for 3 years followed by vertical revenue growth |
| Snap/Instagram |
Neutral (not suitable for EF model) |
Unable to predict winners at day zero; EF would not invest in such companies |
| Facebook/Google |
Risk warning (talent constraint) |
Cannot afford to compete against 100 best ML engineers; Project Maven terminated due to employee protests |
| DeepMind |
Bullish (AI talent hub) |
Demis Hassabis is one of EF's investors |
Judgments Worth Remembering
1. “Every era has its own ‘technological ambition’—literacy, military command, finance, tech entrepreneurship. Today, Napoleon would be ‘green with envy’ at Zuckerberg’s reach.” (Matt Clifford) — Support: The historical arc from Thomas Wolsey (butcher’s son → Cardinal) 1,000 years ago to Napoleon, and then to today’s tech founders, shows leverage growing exponentially.
2. “Traditional VCs say ‘no good team is inefficient,’ we say ‘no efficient team is bad’—so we encourage fast team formation and fast pivots, letting bad teams naturally dissolve.” (Matt Clifford) — Support: EF’s internal co-founder breakup rate is only 10% (industry average ~20%), stemming from a culture of “default no collaboration.”
3. “Every time I added my own ‘signal’ beyond the data, I was wrong. It’s hard to beat 100 consecutive data points—they let you see the shape of the curve, not just a single point.” (Matt Clifford) — Support: The Acurix case—EF’s data ranking was correct, Clifford’s intuition was wrong, and Acurix eventually went from a £275,000 raise to a $10 million Series A.
4. “Deep tech is a natural ally for talent investing—it solves the adverse selection problem. If you’re a world-class CEO wanting to do e-commerce, you don’t need EF; but if you want to do protein discovery, you definitely do.” (Matt Clifford) — Support: Deep tech requires a world-class CTO (cannot be found quickly), is capital-intensive (cannot build an MVP fast), and has well-defined problems (can be judged in advance).
5. “Seed round valuations in Singapore are 50% of those in London, but deep tech companies are global winners. If you do a deep tech seed fund in Southeast Asia and write checks within a week, you’ll win every round.” (Matt Clifford) — Support: EF data shows geography is the biggest driver of seed round pricing, but the difference in exit value is overestimated; speed is the VC quality founders value most.
6. “AI is the first strategic technology where the private sector has an absolute talent advantage. The NSA monopolized cryptography talent 20 years ago; today, Google hires most of the top ML PhDs.” (Matt Clifford) — Support: Google/OpenAI/DeepMind control global AI talent; Project Maven was terminated due to employee protests; no similar constraints exist in China.
7. “Talent is now the most effective constraint for big tech companies—they can afford to confront the U.S. government, but they cannot afford to confront 100 of their best ML engineers.” (Matt Clifford) — Support: Employee protests at Google/Facebook changed corporate behavior; Chinese officials’ response to “employee protests” is “they won’t last long.”
8. “EF’s pricing power doesn’t come from being ‘cheaper than the market,’ but from ‘the market not existing’—we are the only institution investing on Day Zero, capturing the value transformation from ‘0 to 1,’ not capital returns.” (Matt Clifford) — Support: EF generates over 100 interactions with each founder within 100 days, and investment decisions are based on the data generation process rather than personal judgment; EF manages $200 million in assets and has 100 employees.