The Power Law and What It Forces
Returns to venture investments follow a power-law distribution: a small number of extreme outliers carry the whole. From that single empirical fact, the moral pattern of the venture system — the deal-flow imperative, the rejection of the merely-good, the founder selection criteria, the recruitment narrative — falls out as a structural consequence, not a choice anyone made. The piece sets out the chain in five steps and then asks what would change the pattern if the math changed.
The structural argument: why the moral pattern of the venture system is not chosen, and what would change about the pattern if the math changed. Roughly 1,800 words. Citations included; written for readers who want to understand the mechanism rather than read about it.
The distribution
Returns to venture investments do not follow a normal distribution. They follow a power-law distribution, in which a small number of extreme outliers carry the whole. The common rule of thumb in the industry — documented across multiple fund-level studies (Kauffman Foundation 2012, Cambridge Associates pooled return data, Horsley Bridge sample analysis) — is that approximately half of investments return less than capital invested, roughly 30% return between one and three times capital, perhaps 15% return between three and ten times, and a small number, often fewer than 5%, return ten times or more. Within that small number, an even smaller number return one hundred times or more. The fund-level returns to LPs depend almost entirely on whether the fund hit any of those rare extreme outcomes [STRONG].
The structural consequence: a typical venture fund of, say, twenty-five investments needs at least one extreme outlier to return the fund several times over. Without that outlier, the fund returns capital or loses money, regardless of how the other twenty-four did. Hitting the outlier is what fund economics require. Everything else — the working environment, the recruitment messaging, the founder selection criteria, the post-investment pressure, the rejection of modest success — follows from this.
What the math forces, in five steps
One. Funds must take many bets. If a single outlier in twenty-five is what makes the fund work, taking only ten bets is a worse strategy than taking twenty-five. This drives the deal-flow imperative: the fund needs to see thousands of opportunities, fund dozens, and place its outlier-finding bet across as wide a population as the partner's time will allow.
Two. Funds must reject the merely-good. A company that will plausibly return three times capital is good for the founders, good for early employees, and good for the LP capital allocated to it on its own terms. But it cannot be the outlier the fund needs. So the merely-good company is, from the fund's perspective, a worse use of partner attention than the speculative company that might return one hundred times. This produces the well-documented behavioural pattern in which VCs systematically reject founders whose plans look reliably profitable in favour of founders whose plans look implausibly large. The fund is not making a mistake. It is responding to its own incentives correctly.
Three. Founders must believe their company is the outlier. A founder pitching a fund must persuade the fund that the company is in the upper tail of the distribution, not the middle. A founder who acknowledges, accurately, that their company is most likely to be one of the twenty-four-out-of-twenty-five returns less than capital invested will not get funded. So founders are selected for confidence in their own outlier-ness. The selection mechanism rewards over-estimation of personal probability of success, on purpose, because over-estimation is what makes the pitch work.
Four. The recruitment narrative must amplify the outlier-ness signal. For the population of founders who pitch funds to be large enough to find outliers in, the broader recruitment environment around the venture system must persuade many people that they could be the outlier. This is where the cultural infrastructure — the founder hero narrative, the conference keynote, the visible success story, the silence around the failures — does its structural work. The cultural infrastructure is not separate from fund economics. It is the part of fund economics that operates on the population of prospective founders, before any fund partner ever meets any of them.
Five. The system must metabolise failure as ordinary portfolio churn. Twenty-four out of twenty-five companies returning less than capital invested is not a tragedy from the fund's perspective; it is the expected outcome distribution. From the founders' perspective each of those twenty-four is a personal catastrophe of significant magnitude — the years of work, the financial loss, the social cost of public failure, the welfare cost on the people who were employed, the secondary effects on the founders' families. The system has no mechanism for treating these costs as costs. The fund books the loss; the founder absorbs the rest. The asymmetry of who absorbs the cost is structural, not personal.
What this explains
The five-step chain explains, without needing to invoke individual bad behaviour by anyone, the empirical pattern documented elsewhere in the publication: most founders fail (Hall and Woodward 2010), most founders over-estimate their probability of succeeding (Cooper et al. 1988), the founder population shows elevated mental-health diagnoses (Freeman et al. 2019), the demographic distribution of who gets funded is narrow (Crunchbase / PitchBook / Diversity VC), and the recruitment narrative is consistently and systematically optimistic relative to the base rates.
None of this requires a conspiracy. None of it requires VCs to be lying to founders. The mechanism produces the pattern automatically because the math at the top of the funnel forces the behaviour all the way down. Each individual decision in the chain — the partner deciding to take many bets, the partner rejecting the merely-good, the founder pitching outlier potential, the recruitment material featuring survivors, the system metabolising failure as portfolio churn — is locally rational. The aggregate of locally-rational decisions produces the pattern that the publication documents.
This is the version of the argument that does not require villains. It also does not require victims, in the sense of people deceived against their will. The information about the base rates is publicly available. The papers are real. The data is real. What is happening is that the recruitment environment, doing its structural work, is more present in the daily life of a prospective founder than the academic literature is. The asymmetry of attention, not the asymmetry of available facts, is what produces the calibration drift.
What would change the pattern
The structural argument lets us name the things that would change the pattern, in order of how realistic each is.
If fund economics changed. If venture funds had longer time horizons, modestly-sized companies that returned three to five times capital could be acceptable outcomes rather than disappointing ones. Some emerging fund structures — longer-duration evergreen vehicles, small-fund models like Calm Company Fund or Earnest Capital, certain forms of revenue-based financing — have experimented with this. The empirical record on these alternatives is too thin to be confident, but the structural argument is straightforward: change the math at the top, and the pressure all the way down loosens.
If the LP base changed. Pension funds and university endowments allocating to venture have specific return-rate expectations that drive fund behaviour. State-aligned capital, sovereign wealth funds, or mission-aligned LP capital with different expectations could fund a different shape of company-building. The European venture experience, with more state-anchored LP capital and somewhat slower exit dynamics, is sometimes pointed to as a partial natural experiment. The natural experiment is not clean — many other variables differ — but the directional difference in fund behaviour is real.
If founder welfare became a measurable cost on the fund's books. Currently founder welfare is an externality. The fund books financial returns; the founders absorb the welfare cost; no number on the fund's P&L captures the second. If GP-LP-founder governance norms or investment terms made founder welfare partly a fund cost — mental-health insurance, structured rest, founder pay floors, governance protection of non-public-failure pivots — the fund would internalise some of the cost. This is not a current practice; it is a structural option that some recent small-fund experiments are exploring.
If the recruitment narrative carried the base rates. If accelerator marketing, founder-podcast intros, and conference keynotes routinely surfaced Hall-Woodward, Cooper et al., and Freeman et al. alongside the success stories, the calibration drift in the prospective-founder population would be smaller. The system would still need many bets to find outliers, but the bets would come from a population entering with calibrated expectations rather than over-estimated ones. The structural argument suggests this would reduce the population of attempts, which would reduce the search engine's output, which the system has structural reasons not to want. The pressure toward the current asymmetry is, in this sense, endogenous.
The decision the math forces, by reader
If you are a fund partner, the math gives you no good reason to fund the merely-good company over the speculative one. Your incentives are aligned with finding outliers; the merely-good company is the population-level mistake. Acting against the math — longer fund cycles, founder welfare provisions, governance protecting non-public-failure pivots — is a choice you can make. The math does not require you to make it. Whether you do it anyway is a matter of what you think your role inside the system is for.
If you are an LP, the asymmetry between fund returns and founder welfare is invisible on your line in the spreadsheet but real on the ground. Your allocation choices set the math the funds you back must follow. Different LP expectations — longer time horizons, modestly-sized acceptable outcomes, structural founder protections in the fund mandate — would produce different fund behaviour. The funds will respond to whatever you ask them to optimise. What you ask them to optimise is the lever you actually hold.
If you are a founder, the math explains why every interaction you will have with the venture system pulls you toward outlier-shaped behaviour, even when outlier-shaped behaviour is not the right call for your specific company. The pressure is not coming from any individual VC making any individual decision. It is coming from a structure four steps deep, and the structure does not care about your specific case. Knowing this lets you treat the pressure as information about the system rather than information about your company. The information is real and the pressure is real. The conclusion "my company should look like an outlier" is not.
If you are a policymaker, the structural argument tells you which levers actually move the pattern. Capital-gains treatment shifts marginal economics within the existing structure. Fund-structure regulation, LP expectations regimes, and governance norms shift the structure itself. The latter is harder; it is also the only kind of intervention that changes the moral pattern rather than the marginal allocation. Whether it is worth doing is a separate question. But the levers that look most consequential at first glance are usually the ones that change the marginal economics; the levers that actually change the pattern are the ones that change the math.
Where to go next
The headline argument with both halves: Venture Capital Is Good for Society and Bad for Most Founders.
The recruitment-narrative argument as its own piece: For prospective founders — what the recruitment narrative does not say.
The empirical detail on failure rates, mental-health prevalence, ADHD, and the demographic distribution: The reality of being a founder.
The full-length analytical version with seven evaluative frames in parallel: VC: most fail, most suffer, some win lots.