Eighty years later, the startup world runs on the same mistake. We study the survivors — the founders who built billion-dollar companies, the executives who turned around failing divisions, the operators who timed their market perfectly — and we extract "lessons" from their stories as though those lessons caused the outcomes. The people who did the same things and failed are not on the podcast circuit. They are not writing bestsellers. They are not keynoting conferences. They are the bombers that did not come back, and nobody is studying them.
The problem is not anecdotal
Daniel Lerner at IE Business School and colleagues at Baylor University and Virginia Tech published research through the Babson College Frontiers of Entrepreneurship series quantifying exactly how badly survivorship bias distorts entrepreneurship data. Using a dataset that comprehensively captured every firm in an industry — not just the ones that survived — they found that entry was understated by 40% and average firm longevity was overstated by 152% when failed firms were excluded. The distortion was not subtle. The data told a fundamentally different story when the dead were included.
This is not a niche academic complaint. It means that every study, every analysis, every "lessons from successful companies" article that does not include the companies that tried the same things and failed is producing conclusions that may be not just incomplete but actively wrong. The strategies that look like causes of success may be equally common among failures. The habits that look like differentiators may be table stakes — things everyone does, with outcomes determined by variables the study never measured because it never looked at the companies where those variables produced a different result.
Ethan Mollick, a professor at Wharton, has written extensively about what he calls "superstitious learning" — the phenomenon where people learn the wrong lessons by focusing on symbols of success rather than the harder-to-identify systems that actually drove outcomes. In a paper analyzing survivorship bias in startup contexts, researchers showed that when people were told unicorn founders were disproportionately college dropouts (which is not true), they shifted their expectations dramatically. People shown examples of successful dropouts were 55 percentage points less likely to bet on graduates than people shown examples of successful graduates. The bias did not just distort understanding — it reversed it.
Mollick's conclusion is blunt: in fields where luck is a major factor in success, trying to learn from the best-performing companies is "not only fruitless, it is destructive." The best learning comes from companies that performed well but not exceptionally — companies more likely to have embraced less risk and more likely to be genuinely skilled rather than genuinely lucky.
What the success stories leave out
Every startup mythology contains a version of the same story: the founder who dropped out, maxed out their credit cards, slept in the office, ignored conventional wisdom, and built something worth billions. The implicit lesson is that these behaviors — the risk-taking, the sacrifice, the contrarianism — caused the success. But the research says otherwise.
The mean age of a startup founder is 42, not 22. The founders most likely to build high-growth companies are between 45 and 59, not fresh out of college. Flat organizational structures, often celebrated as innovative, tend to hurt companies as they grow — hiring good middle managers is one of the strongest predictors of startup success. These findings, drawn from research by Pierre Azoulay and colleagues published in the American Economic Review, contradict almost everything the survivor-driven narrative teaches. But they come from datasets that include companies that failed, not just companies that keynote at conferences.
Harvard professors Francesca Gino and Gary Pisano have documented what they call the three learning traps of success: attribution error (crediting your strategy when the market conditions did the work), overconfidence bias (believing your judgment is better than it is because it happened to produce a good outcome), and the failure to learn (because success does not force the kind of analysis that failure does). Their research shows that leaders systematically learn less from success than from failure — which means the entire industry of "learn from successful founders" is building on the weakest possible foundation.
The advice paradox compounds this. When a successful founder gives advice, they are not lying. They genuinely believe that their specific decisions caused their outcomes. But they are a sample size of one, drawn from the surviving population, with no control group. They cannot know — and often have never considered — how many other founders made the same decisions and got a different result. Confirmation bias means they remember the decisions that look prescient in hindsight and forget or reframe the ones that do not fit the narrative.
The 90% you never hear from
Roughly 90% of startups fail. This is not a contested figure — CB Insights, Failory, the Bureau of Labor Statistics, and multiple academic databases converge on approximately the same number. What is less discussed is what this means for the advice economy.
If 90% of startups fail and the advice ecosystem draws almost exclusively from the 10% that succeed, then every "lesson" extracted from that ecosystem is drawn from a population that represents — at best — one-tenth of the actual experience base. The other nine-tenths worked just as hard, made similar bets, followed similar playbooks, and got a different result. They had the same morning routines, the same hustle, the same conviction. The difference was often timing, market conditions, team composition, capitalization, or luck — variables that do not make for compelling podcast episodes.
An analysis of 150+ startup post-mortems found that the most common causes of failure were painfully ordinary: poor market timing, hiring the wrong team, weak marketing, misaligned founder dynamics, and running out of cash due to preventable overspending. These are not exotic. They are not failures of vision or courage. They are operational failures that happen to the exact type of founder who gets celebrated when the same operational dice land differently.
The Babson research makes this quantitatively precise. When the researchers excluded failed firms from their dataset — mimicking what every "lessons from success" study does — the resulting picture of the industry was so distorted that it "dramatically mischaracterized industry dynamics and firm-level outcomes." Entry rates, survival rates, and performance metrics all told a fundamentally misleading story. The data was not slightly off. It was structurally wrong.
Why this matters more than you think
The practical damage of survivorship bias is not that founders read the wrong books. It is that survivorship bias distorts decision-making at the exact moments when accurate pattern recognition matters most.
When a founder decides whether to pivot, they compare their situation to the successful founders they have studied — the ones who "stuck with it" and eventually won. But for every founder who persisted through a downturn and came out the other side, there are dozens who persisted through a downturn and ran out of runway. The sunk cost fallacy and survivorship bias reinforce each other here: the founder keeps investing because they have already invested too much to stop, and they justify the decision by pointing to successful founders who also kept going. The missing data — all the founders who also kept going and failed — is invisible.
When a founder evaluates their own performance, they benchmark against the survivors. This creates what psychologists call an upward social comparison with an unrepresentative sample. Social comparison theory predicts that comparing yourself to people who have achieved more produces either motivation or demoralization — but when the comparison group is systematically skewed toward the luckiest outcomes, both responses are miscalibrated. The founder either feels inadequate (because they are comparing themselves to statistically anomalous results) or feels confident that their trajectory matches the success pattern (because they are matching against a pattern that includes luck they cannot replicate).
Founders make worse decisions alone in general, but survivorship bias makes solitary decision-making specifically dangerous because the reference material available to a solo founder is almost entirely survivor-generated. Books, podcasts, case studies, conference talks, Twitter threads — the entire advice supply chain is filtered through the same bias. A founder working alone has no mechanism to introduce the missing data.
Where the corrective actually lives
Abraham Wald solved the bomber problem not by studying the survivors more carefully but by asking a different question. Instead of "where are the bullet holes?" he asked "where are the bullet holes not?" The corrective was not better analysis of the available data. It was recognizing that the available data was the wrong data.
For founders, the equivalent corrective is exposure to the full distribution of outcomes — not just the ones that made it. And the most practical way to get that exposure is through peers who are in the middle of their own uncertain journeys, not through mentors and speakers who have already emerged on the other side of theirs.
A peer group composed of founders at similar stages introduces what survivorship bias removes: the base rate. In a group of eight founders, statistically six or seven of them will not build a company that reaches the outcomes celebrated in the advice economy. This is not pessimism. It is math. And it is exactly the math that the podcast-and-keynote ecosystem hides.
When a founder sits in a hot seat with five other founders, they are not hearing from survivors. They are hearing from people who are navigating the same uncertainty, making the same kinds of bets, and discovering the same kinds of problems — in real time. The advice is grounded in current conditions, not in a retroactively cleaned-up narrative. The feedback is forward-looking, not backward-rationalizing. And the implicit comparison — "how am I doing relative to others at my stage?" — is calibrated against a representative sample rather than a skewed one.
Your network is too big to help you with this problem because large networks amplify survivorship bias. The people with the loudest voices in any professional network are the ones who succeeded — they have the platform, the credibility, and the incentive to share their story. A tight peer group of five to eight founders is not a network. It is a dataset. A small, honest, representative dataset of what building a company actually looks like right now, not what it looked like in retrospect for the 10% who made it.
The question Wald would ask
Wald's insight was not complicated. It was a refusal to accept the available data as the whole picture. He looked at the bombers that came back and asked: what about the ones that did not?
The founder version of this question is: what about the people who did exactly what I am doing and it did not work? Where are they? What happened? What did they see that I am not seeing? Pluralistic ignorance ensures that most founders believe they are the only ones struggling, because the visible evidence — the survivors — suggests that struggle is temporary and victory is inevitable for those who persist.
The armor goes where the bullet holes are not. The learning lives in the failures nobody writes about. And accountability — real accountability, the kind that comes from people who will challenge your narrative rather than affirm it — is the structural mechanism that forces a founder to see the whole picture rather than just the surviving slice of it.
You are reading this article because you are still here. You are one of the planes that came back. That is good. But it does not mean you know where the danger is. The planes that came back never do. The danger lives in the places you have not been hit yet — and the only way to see those places is through the eyes of people who are not looking at your bullet holes. They are looking at the empty spaces.
That is what a peer group does. Not better advice from better survivors. Better questions from people who are still in the air.