Investors aren't afraid you'll lose money
They're afraid you'll do everything right and still be small

The mechanism is the structure of the fund, not the personality of the investor. His money sits in a fund with a clock on it, his result is judged on the whole portfolio, and that portfolio's return is dominated by a couple of winners. That makes his two errors asymmetric: missing a winner costs one opportunity, while backing a company that quietly plateaus ties up capital and a board seat for a decade. The rational response is to miss a great deal rather than fund something that can't be enormous. The steadier you sound, the more certain he is that the ceiling is already set; the wilder you sound with no evidence, the more certain he is that you don't know where you're going. What survives is the third thing: credible today, open-ended at the top.

• Industry data · returns follow a power law: Correlation Ventures' analysis of roughly 21,000 US venture financings from 2004–2013 found about 65% of deals returned less than the capital invested and only about 4% returned more than 10x. Strength: proprietary industry data, not peer-reviewed, dependent on its own data sources — fine as an order of magnitude, not as a precise constant.
• Large survey · how hard they filter: Gompers, Gornall, Kaplan & Strebulaev (2020, Journal of Financial Economics), "How do venture capitalists make decisions?", surveyed nearly 900 institutional investors: for each deal they close, firms consider around 100 opportunities on average. Strength: large-sample self-report — it tells you how they believe they decide, not causation.
• Boundary · this arithmetic belongs to venture capital alone: a bank asks whether you can repay, a customer asks whether you can deliver, a revenue-based lender asks whether cash flow is steady. The same company must tell genuinely different stories to different pools of money — not because anyone is lying, but because they are buying different things.

Boundaries and failure modes: power-law logic applies to money that needs fund-level returns. Individual angels, strategic investors and corporate funds optimise for other things (learning, supply-chain fit), and forcing a "this will be a hundred-billion market" frame onto them just blurs the pitch. Also, don't turn "open-ended" into fabrication: every large number you offer must reduce back to a small number someone can check today, or it shatters in the first diligence call.

Two mechanisms, both about processing bandwidth. One, redundancy: when the same words are on the screen and in your mouth at once, two streams compete for one channel and comprehension drops — so the slide keeps one sentence, one image or one number, and you say the rest. Two, thin slices: people form surprisingly stable judgements within seconds, and later information is largely recruited to confirm them. That's not an instruction to perform confidence; it's an instruction to put the highest-information items first — your positioning, and one number someone else could check.

• Peer-reviewed · redundancy hurts comprehension: among the multimedia learning principles summarised by Mayer & Moreno (2003, Educational Psychologist), narration plus identical on-screen text produces worse learning than narration with graphics alone. It replicates across experiments; the mechanism is channel competition in working memory.
• Peer-reviewed · thin slices are fast and sticky: Ambady & Rosenthal (1993, JPSP) found ratings from thirty-second silent video clips of teachers correlated strongly with formal end-of-term student evaluations. Stated honestly: thin slices predict how others will rate you, which is not the same as predicting real ability — that's exactly what makes the effect both useful and dangerous.
• Heuristic (weak evidence): Kawasaki's "10/20/30" is widely repeated but it is practitioner observation, not a controlled experiment — fine as a discipline device, not as a law to cite.

Boundary: less is not always more. The version that circulates internally, and the material used in diligence, needs to be complete and checkable — that one can be thick; the version you speak over must be thin. Keep both, and don't let either impersonate the other. One more thing: what usually kills a deck isn't the page count, it's that you don't say what you actually do until page seven.

The mechanism is costly signalling under information asymmetry: when the receiver can't verify quality directly, the only things that carry information are behaviours that would be more expensive for a pretender. Passion, determination and conviction about a direction cost nothing, so in equilibrium they get ignored entirely — not because investors are cold, but because those words are mathematically empty. A cost already paid — a thing built, money received, a repeat purchase, a specific hole you fell into — can't be faked for free, so it gets taken seriously. This is also why "team" carries the most weight at the early stage: the product will change and the market will change, but what this group has already spent on the problem is the hardest thing to fabricate.

• Hard evidence (theoretical foundation) · signals must be costly to work: Spence (1973, Quarterly Journal of Economics), "Job Market Signaling": in markets with asymmetric information, only signals that are more expensive for low-quality senders can sustain a separating equilibrium. The work earned the 2001 Nobel Prize in economics and has been extended and tested repeatedly.
• Large survey · the team dominates early: in Gompers et al. (2020, JFE), a survey of nearly 900 investors, about 47% named the team as the most important factor, and they attributed the success or failure of portfolio companies primarily to the team rather than to the business model. Strength: self-report, which may overstate their insight into their own process.
• Peer-reviewed (qualitative) · the early stage is a bet on people: Huang & Pearce (2015, Administrative Science Quarterly) found angel investors explicitly rely on "gut feel" as a decision tool at the earliest stage, where hard data barely exists. Stated honestly: this is interview- and observation-based work describing what decision-making looks like — it is not an endorsement that intuition is accurate.

Boundary: costly signals get faked too — inflated user counts, trials booked as paying customers, a friend's order presented as market validation. Diligence surfaces almost all of it, and what you lose isn't a round, it's your name in a very small industry. There's also an unfair reality worth holding: "costly" is relative to personal resources — burning six months of savings on a prototype costs far more if you support a family. That doesn't change the mechanism, but it's worth remembering when you judge other people by it.

Three mechanisms. One, ambiguity is worse than risk: people react far more negatively to unknown odds than to known bad odds, which is why "I don't know yet, here's how I'll find out and by when" beats both a vague dodge and an invented number — it converts ambiguity into risk. Two, two-sided arguments: naming a weakness and answering it persuades better than a one-sided case, but the refutation is mandatory — conceding without answering lands below saying nothing at all. Three, frame stickiness: the frame of the question becomes the frame of the answer, and defensive questions answered defensively turn the whole meeting into a risk inventory.

• Hard evidence (brain mechanism) · ambiguity and risk don't run on the same system: Hsu, Bhatt, Adolphs, Tranel & Camerer (2005, Science) found that ambiguity (unknown probabilities) produced stronger activation in the amygdala and orbitofrontal cortex, while known-probability risk produced a stronger response in the striatum; patients with orbitofrontal damage barely distinguished the two. fMRI plus lesion patients — hard. Turning "I don't know" into "a known uncertainty with a plan to resolve it" is exactly that conversion.
• Meta-analysis · conceding only helps if you refute: O'Keefe (1999), a meta-analytic review of one-sided versus two-sided messages: two-sided messages that acknowledge and then refute the opposing point are more persuasive than one-sided ones, while two-sided messages that acknowledge without refuting perform worse than one-sided. Meta-analytic and directionally stable. It hands you the shape of the script directly: concede, then answer — neither half is optional.
• Observational plus experimental · framing predicts how much you raise: Kanze, Huang, Conley & Higgins (2018, Academy of Management Journal) coded live pitch Q&A at a startup competition: investors asked male founders more "how will you win" (promotion) questions and female founders more "how will you avoid losing" (prevention) questions; founders who received prevention questions and answered inside that frame went on to raise dramatically less — an order-of-magnitude gap; accompanying experiments showed that reframing a prevention question into a promotion answer improves outcomes. Stated honestly: the field data is observational and open to confounds, while the reframing effect has experimental support. This does not put the responsibility for a structural bias on the person being questioned — it gives someone inside that system something usable today, while the system is slower to change.

Boundary: reframing is not evasion. Dodging a sharp question costs far more than answering it imperfectly — dodge once and he'll assume every corner he didn't inspect hides something similar. Don't perform modesty either ("our biggest weakness is that we work too hard"); that's worse than conceding nothing. The shape here is the same one used for handling objections: catch it, answer it, then carry the conversation back to where you were going.