Why Startups Fail: Execution Risk Beats Competition

Paul Graham posted a line this week that founders have been passing around like a hot coal: you are far more dangerous to your startup than your competitors are. A hundred times more startups die from poor execution by their founders than are killed by competitors. When someone pushed back that the advice sounded obvious, he doubled down. The reason it isn’t vacuous, he said, is that founders don’t realize it. So they worry more about the much smaller risk posed by competitors.

I’ve lived both sides of that sentence. At CommonFloor we spent real energy in the early years watching rivals raise money, copy features, and out-spend us on marketing. Almost none of that energy changed a single decision that mattered. What nearly killed us at various points was always internal: a hiring mistake, a product bet we held too long, a quarter where we scaled spend ahead of revenue truth. The competitors we feared most are mostly gone now, and not one of them took us down on their way out.

So this essay is the durable version of that hot coal. Not a hot take on one post, but the full argument: what the mortality data actually says about why startups fail, why founder attention gets allocated backwards, and a set of tools to point your worry at the things that are actually trying to kill you.

Fair warning: the data is uncomfortable. The killer almost always has your face.

Table of Contents

• You Are Watching the Wrong Killer
• The Failure Ledger: Where Startups Actually Die
• Failure Mode 1: Building Something Nobody Wants
• Failure Mode 2: Scaling Before the Truth Arrives
• Failure Mode 3: Economics That Never Close
• Failure Mode 4: The Founder System Fails First
• When Competition Actually Kills
• The Contrarian Take: Competitor Research Is Mostly Procrastination
• The Monday Morning Execution Audit
• FAQ

You Are Watching the Wrong Killer

Ask a room of founders what keeps them up at night and you’ll hear the same answers: a better-funded competitor, a big incumbent waking up, someone shipping their roadmap first. Ask the data what actually killed the last few hundred startups and you get a completely different list.

CB Insights ran the autopsy on 431 VC-backed companies that shut down from 2023 onward. The headline number is that 70% ran out of capital, but that’s the cause of death the way “cardiac arrest” is a cause of death. Everything ends with the money stopping. The root causes underneath are the story: 43% never found product-market fit, 29% mistimed their market, 19% had unit economics that never worked. Getting outcompeted shows up at just 19%, and as we’ll see, even that number is softer than it looks.

This isn’t a one-report fluke. An earlier CB Insights analysis of 101 founder post-mortems found 42% citing “no market need” as the top reason. A separate study of 200 post-mortems by the research group Autopsy found that being outcompeted was essentially absent among bootstrapped companies and those that raised under $1 million. The pattern holds across datasets, across cycles, and across a decade of these reports: startups overwhelmingly die of internal causes.

The volume makes this urgent. Carta counted 966 US startup shutdowns in 2024, up 25.6% from 769 the year before. Enterprise SaaS alone accounted for 32% of them. That’s a lot of post-mortems, and the striking thing about reading them is how rarely a competitor appears in the death scene at all. The companies describe their own hiring, their own burn, their own product misses, their own founder fights.

And there’s a quieter statistic inside the CB Insights work that I think about a lot. Of the dead companies with a full year of health-score data before shutdown, 72% saw their score decline through that final year, dropping 15% on average. The decline was visible. The instruments were flashing. Death by execution is rarely sudden; it’s a slow leak that founders normalize while they refresh a competitor’s launch page.

The problem, then, isn’t that founders don’t worry enough. It’s that founder worry is a portfolio, and most of us have it allocated backwards.

The Failure Ledger: Where Startups Actually Die

Here’s the model I use. Every startup death gets booked to one of two accounts. The external account: killed by something outside the building, like a competitor, a platform shift, or a regulation. The internal account: killed by something inside the building, like a product nobody wanted, spending ahead of evidence, economics that never closed, or the founding team breaking. I call this the Failure Ledger, and the entire argument of this essay is that the ledger is wildly lopsided and founder attention is allocated as if it were balanced.

Three things jump out of the ledger when you stare at it.

First, capital is never the real entry. “Ran out of money” is how every startup death certificate reads, which is exactly why it tells you nothing. Money runs out because of an upstream failure that was bookable months or years earlier. Treat the 70% as the smoke, not the fire.

Second, the internal account dwarfs the external one. Add up the self-inflicted root causes and you get a multiple of the outcompeted line. And the multiple understates it, because post-mortems are written by founders, and founders flatter themselves. “We got outcompeted” is a much more comfortable sentence than “I kept funding a product the market had already rejected.” When the Autopsy researchers dug into who actually cites competition, it was concentrated among companies that had raised eight figures, the exact cohort with the most incentive to blame something external. Behavioral and interpersonal failures, meanwhile, are systematically underreported because nobody wants to write “my co-founder and I stopped speaking” in a public farewell post.

Third, every line in the internal account was controllable. Not easy, but controllable. Nothing on that side of the ledger required a rival to do anything. That’s the actual meaning of the hundred-to-one framing: not that competitors never matter, but that the expected damage from your own execution dwarfs the expected damage from theirs, and only one of those is fully in your hands.

Y Combinator has compressed this into one of its essential pieces of advice: startups die of suicide, not murder. The phrasing is blunt because the bias it’s fighting is strong. Competition is visible, dramatic, and external, so your brain treats it as the predator. Execution decay is invisible, boring, and self-implicating, so your brain files it under “fine.” The Failure Ledger is just a tool for overriding that bias with base rates.

A useful way to operationalize the ledger is to treat founder worry as a portfolio with positions. You hold worry positions in competition, in product, in burn, in team. Like any portfolio, returns come from allocation, and the allocation most founders run is inverted relative to the risk: heavily overweight the 19% line they can’t control, badly underweight the 43% and 74% lines they can. Rebalancing costs nothing. Moving attention from the external account to the internal one is the only free alpha in startup land, and it improves the expected value of every week you work.

The rest of this essay walks the internal account line by line: the four failure modes that actually fill the ledger, how each one disguises itself while it’s happening, and the specific checks that catch it early. Then we’ll give competition its honest due, because the external account isn’t zero, and pretending it is would be its own execution failure.

Failure Mode 1: Building Something Nobody Wants

The 43% line. The biggest single entry in the ledger, and the one Paul Graham has called the number one cause of startup death for nearly two decades: making something no one wants.

What’s strange is that no founder believes this applies to them at the time. PMF failure almost never feels like building something nobody wants. It feels like building something people are about to want. The signals are always six weeks away. The pilot customer is always about to convert. The redesign is always going to unlock activation. I’ve seen founders, smart ones, run eighteen months on “about to.”

The mechanics are worth understanding because they’re so repeatable. You launch. Usage is polite but shallow. Instead of treating shallow usage as the market’s answer, you treat it as a messaging problem, then a feature-gap problem, then a sales problem. Each reframe buys another quarter of burn. The CB Insights health-score data shows exactly this shape: 72% of dead companies were measurably declining for a full year before the end. The market had answered. The founders were still negotiating.

Quibi is the canonical large-scale version. $1.75 billion raised, a founding team with legendary resumes, and a thesis (premium short-form video for in-between moments) that the market had never validated. Six months after launch it had roughly 500,000 paying subscribers against targets that assumed millions, and it shut down having torched about $1.4 billion. No competitor did that. YouTube and TikTok didn’t out-execute Quibi; Quibi paid $1.75 billion to learn the demand wasn’t there, when a fraction of that would have surfaced the same answer.

The defense is not complicated, but it is emotionally expensive: ship the smallest testable version, charge real money early, and treat retention as the only PMF signal that can’t lie to you. Acquisition can be bought and hype can be manufactured, but nobody keeps using a product weekly out of politeness. I walked through the full validation sequence in how to validate a startup idea in 48 hours, and the core of it is exactly this: compress the time between your belief and the market’s verdict, because every week of gap is burn spent on a question that already has an answer.

One more tell, because it’s the cheapest diagnostic available: listen to how you describe traction. If your sentences are about inputs (features shipped, meetings taken, pipeline built) rather than outcomes (retention, revenue, referrals), some part of you already knows what the outcome numbers say.

One more thing has changed since the classic post-mortem studies, and it raises the stakes on this failure mode rather than lowering them. Building is no longer the bottleneck. With AI coding agents, a working product takes a weekend, which means markets now fill with plausible products faster than demand can sort them. When building was expensive, the cost of building the wrong thing was capped by how slowly you could build. Now you can build the wrong thing at unprecedented speed, ship six variations of it, and mistake your own velocity for traction. Cheap building makes validation the entire game: the scarce skill isn’t shipping anymore, it’s knowing what deserves to be shipped.

Failure Mode 2: Scaling Before the Truth Arrives

The 29% “bad timing” line in the ledger is usually read as “the market wasn’t ready.” Read the actual post-mortems and it’s more often the inverse: the company scaled as if a truth had arrived when it hadn’t. Startup Genome’s analysis of 3,200 high-growth companies put a hard number on it that still shocks people: 74% of high-growth internet startup failures trace to premature scaling. And in their dataset, not a single startup that scaled prematurely passed the 100,000-user mark. Not one.

Premature scaling means spending like a company one stage ahead of your evidence. Hiring a sales team before the founder can sell it repeatably. Buying growth before retention supports it. Building enterprise infrastructure before ten enterprises care. It’s the most seductive failure mode because every individual act of it looks like ambition, and ambition is what everyone praised you for at the last board meeting.

Fast is the cleanest modern specimen. The one-click checkout startup raised over $120 million, including a $102 million round led by Stripe, hired hundreds of people including layers of expensive executives, and sponsored everything that moved. Its revenue for the year before it died: roughly $600,000. That is a company buying the costume of a winner before the body existed. When it shut down in April 2022, no competitor had beaten it. Bolt and Shopify’s Shop Pay were out there, sure, but Fast didn’t lose a checkout war. It spent itself to death while the unit of value, merchants actually adopting and transacting, never materialized at scale.

Here’s the supporting visual I wish someone had taped to my monitor in my scaling years. It’s the worry inversion in one table: where founder attention typically goes versus where the bodies actually are.

The discipline that beats premature scaling is sequencing, not frugality. Spend aggressively on whatever produces evidence (prototypes, customer conversations, paid pilots) and stingily on everything that assumes the evidence already exists (headcount, brand, infrastructure). I run my own version of this as a solo founder with AI doing the heavy lifting, which I documented in the solo founder AI operating system: the entire point of that system is that headcount is no longer the only way to buy speed, which makes premature hiring an even less defensible bet than it was a decade ago.

The question that catches this failure mode early is brutally simple: what evidence did we get this month that justifies next month’s spend? If the honest answer is “the same evidence we had last quarter, but we’re more excited now,” you’re scaling on enthusiasm. Enthusiasm is not a balance sheet item.

Failure Mode 3: Economics That Never Close

The 19% unit economics line is the quiet one. It doesn’t make dramatic post-mortems because it’s arithmetic, and arithmetic is embarrassing in a way that market forces aren’t. But it’s the failure mode that’s been growing fastest in relevance, because two structural shifts made bad unit economics easier to hide and faster to kill.

The first shift is that blitzscaling-era instincts outlived the blitzscaling-era capital. A generation of operators learned that negative margins were fine because the next round would be bigger. Carta’s shutdown numbers (966 in 2024, up 25.6% year over year) are in large part that assumption dying on contact with a market where the next round stopped being automatic. The companies didn’t change; the tolerance for arithmetic that never closes did.

The second shift is specific to building with AI, and I see founders walk into it weekly. Inference costs sit in your cost of goods sold, not your R&D line. A SaaS product at 80% gross margin can absorb a lot of sins; an AI product running heavy model calls per user action can quietly sit at 30% or even negative contribution margin per power user, which means growth makes you die faster. I went deep on the cost side of this in what AI reliability actually costs, and the punchline transfers: if you don’t know your cost per successful user outcome, you don’t know whether you have a business or a subsidy program.

What makes broken economics a self-inflicted wound rather than a market condition is that the numbers are always available before the death. Fast knew its revenue. Quibi knew its conversion. The failure is never information; it’s the willingness to act on it, because acting on it usually means shrinking something (the team, the free tier, the growth rate) and shrinking feels like losing. So founders defer the arithmetic and call it optimism.

The check here is one number per business model. If you’re transactional, contribution margin per transaction including all variable costs. If you’re subscription, gross margin after the cost to serve your heaviest-usage decile. If you’re marketplace, take rate minus the incentives you’re spending to fake liquidity. One number, updated monthly, with a single rule attached: the number has to be on a path to positive before you’re allowed to buy growth. Not at positive necessarily, but on a path you can defend with this month’s data rather than next year’s hopes.

Failure Mode 4: The Founder System Fails First

Here’s the entry that barely shows up in public post-mortems and dominates the private ones. Noam Wasserman’s research at Harvard, following roughly 10,000 founders, found that 65% of high-potential startup failures trace to co-founder conflict. Sixty-five percent. Higher than any market factor in any dataset, and it’s the line item nobody writes the farewell blog post about, because “we couldn’t agree on who was CEO” doesn’t flatter anyone.

I’d widen the category beyond co-founder fights, because the underlying asset is the same: the founder system, meaning the decision-making capacity, energy, and alignment of the people steering. Startups are unusual in that the steering system and the cargo are the same people. When a founder burns out, starts avoiding hard conversations, or stops being able to kill their own bad ideas, the company develops the same symptoms as a market failure: slow decisions, drifting roadmap, talent leaving. It gets diagnosed as everything except what it is.

The failure chain usually runs: misalignment, then avoidance, then parallel agendas, then a forcing event (a fundraise, a pivot decision, an acquisition offer) that the relationship can’t survive. Wasserman’s data has a hopeful edge though: founder agreements that pre-decide the hard questions (equity vesting, decision rights, exit conditions) cut the conflict risk by around 44%. The fix is mostly done before the fight, not during it.

For the energy half of the founder system, the failure mode is subtler than burnout headlines suggest. It’s not collapsing at your desk; it’s the slow narrowing where you only have capacity for reactive work, and the proactive work (the customer calls, the strategy reviews, the uncomfortable metric inspections) silently stops. Every execution failure in this essay accelerates when the founder is running on fumes, which is why I treat energy as infrastructure rather than virtue. I wrote up my whole protocol in the operating energy OS for founders, and separately, the ego side of the system, the ability to let a losing idea die before it takes the company with it, in the art of killing your ego.

The audit question for this failure mode: when did you and your co-founder (or you and yourself, solo founders) last disagree about something important, out loud, and resolve it? If the answer is “we never disagree,” that’s not harmony. That’s avoidance compounding at startup interest rates.

When Competition Actually Kills

Now the honest part. The external account isn’t empty, and a framework that says “never think about competitors” is just the original error mirrored. The interesting question is when competition graduates from background noise to actual threat, and the data gives a surprisingly crisp answer.

Remember the Autopsy finding: being outcompeted was cited by 19% of companies that raised $10 million or more, and almost never by bootstrapped companies or sub-$1M raisers. That’s not because bootstrappers are invincible. It’s because competition kills through a specific mechanism: it compresses the time you have to fix your execution, and it punishes capital-intensive positions hardest. A bootstrapped product with 200 paying customers and positive margins can survive a funded rival indefinitely; it just grows slower. A company that raised $50 million on a land-grab thesis can be killed by a competitor, because the competitor invalidates the thesis the valuation was built on, the next round dies, and the burn does the rest. Even then, look at the death certificate closely: the proximate cause was a spend structure that assumed winning. The competitor just removed the assumption.

So here’s the filter I actually use when competitor anxiety spikes. It’s a decision tree with a strong default, and the default is “back to work.”

Notice what the filter does: even in the worst case, where a competitor genuinely invalidates your funding thesis, the action item is internal. You restructure your spend, refocus on a defensible segment, and extend runway. Competition, when it matters, matters by forcing an execution decision. Which means the response to competition is also execution. There is no branch of the tree where “monitor them harder” is the move.

The other legitimate competitive concern is structural rather than tactical: are you building anything that compounds, so that execution today buys defensibility tomorrow? That’s a moat question, and it deserves deliberate quarterly thought rather than daily anxiety. I laid out my approach in the data moat playbook. The healthy cadence: moats quarterly, customers weekly, competitors monthly at most.

The two external threats that deserve real respect

If I’m steelmanning the external account, two entries deserve more respect than the average competitor. The first is platform dependency. When your distribution, your data, or your core functionality lives on someone else’s platform, that platform’s roadmap is a loaded gun pointed at your business, and history is littered with companies that learned this when an API changed terms or a feed algorithm shifted. Platform risk is genuinely external, but notice that the exposure to it was an internal choice: how much of your business you built on rented land was a decision you made, and it can be unwound the same way.

The second is the incumbent shipping your product as a feature. Every founder building on top of foundation models has felt this one: the fear that the next model release simply absorbs your category. It’s a real risk, and the test is the same one from the decision tree: does the incumbent’s move change what your customers need from you? If your entire value was a thin convenience layer, then yes, and the honest conclusion is that you never had a defensible product, which is an execution verdict, not a competitive one. If you own workflow, data, integration depth, or trust that the incumbent can’t replicate quickly, their announcement is marketing for your category. I wrote about engineering that distinction deliberately in how an AI business survives model churn. Either way, the response runs through your own roadmap.

The Contrarian Take: Competitor Research Is Mostly Procrastination

Here’s what I think most people get wrong, including most of the people nodding along with the hundred-to-one line: they treat competitor obsession as a harmless vice. A little market awareness, surely that can’t hurt. I’d argue it actively selects for failure, for a reason almost nobody says out loud.

Watching competitors feels like work but carries no accountability. If you spend the afternoon in a rival’s changelog, you’ve “done strategy.” Nothing you learned can be falsified, nothing you concluded has a deadline, and no metric will ever tell you it was wasted. Compare that with the alternatives: a customer call can reject your roadmap, a pricing test can fail publicly, a retention query can tell you your product isn’t working. Competitor research is the only “strategic” activity with zero personal downside, which is precisely why anxious founders default to it. It’s procrastination wearing a strategy costume.

And it’s worse than neutral, because it pulls your roadmap toward convergence. Founders who watch competitors ship competitor-shaped features. You end up in feature-parity races where the prize for winning is sameness, in a game where differentiation was the only durable advantage on offer. I’ve watched companies copy a rival’s pricing page while their own churn data, which contained the actual answer, sat unread. The deep irony of competitor obsession is that it makes you worse at competing.

There’s also a calibration angle. Fear of competitors is unfalsifiable and therefore unbounded: there is always a scarier rival rumor available if you want one. Execution fears are bounded by data: your retention is a number, your margin is a number, your runway is a number. A founder who moves their anxiety from the unbounded register to the bounded one doesn’t just allocate attention better; they become measurably saner. That skill, keeping your internal probabilities tied to evidence, is trainable, and I wrote the training plan in the founder’s calibration practice.

So my actual position is stronger than “worry less about competitors.” It’s: treat any unscheduled competitor browsing as a symptom. The moment you catch yourself doing it, ask what uncomfortable internal number you’re avoiding. In my experience the answer is always available and always specific. The competitor tab is where founders hide from their own dashboards.

The Monday Morning Execution Audit

Frameworks are decoration unless they change what you do this week. Here’s the audit, five checks mapped to the five ledger lines. Run it Monday morning. Twenty minutes, no meetings required.

Three implementation notes from running versions of this on my own companies.

First, the order matters. Retention before anything else, competitors dead last. The sequence is the point: it rehearses the correct worry allocation until it becomes reflex. If you only adopt one thing from this essay, adopt the ordering.

Second, write the answers down, one line each, same doc every week. The value compounds in the diffs. A single week of “haven’t looked at retention” is noise; three consecutive weeks is the 72%-decline curve starting, and you’ll see it in your own handwriting before any dashboard makes it undeniable. This pairs naturally with a quarterly pre-mortem practice, which is the same instinct pointed forward: name the self-inflicted wound before you inflict it.

Third, the audit only works if bad answers trigger action the same week. A red flag on market need means customer calls go on tomorrow’s calendar. A red flag on spend pacing means the hiring plan gets a second look before the offer goes out. The founders who die with declining health scores for a full year weren’t missing information. They were missing a forcing function. This is yours.

If you’re earlier in the journey and the audit feels heavyweight, the minimum viable version is two questions: did I talk to a customer this week, and did my one margin number move the right way? Those two cover 62 points of the ledger by themselves. And if you find yourself drowning in analysis instead of shipping, that’s its own failure mode, which I covered in from overthinking to over-shipping.

FAQ

What percentage of startups fail because of competition?

Around 19% of failed startups cite being outcompeted, per CB Insights’ analysis of 431 post-mortems, and the Autopsy study found the figure concentrated almost entirely among companies that raised $10 million or more. Among bootstrapped startups, competition is almost never cited as the cause of death. Self-inflicted causes like no market need (43%) dominate every dataset.

What is the number one reason startups fail?

Building something the market doesn’t want. CB Insights found 43% of dead startups lacked product-market fit (an earlier study put “no market need” at 42%). While 70% technically “ran out of money,” that’s the symptom: the money ran out because an upstream execution failure, usually missing market need, was left unfixed.

What is premature scaling and why is it so dangerous?

Premature scaling means spending like a company one stage ahead of your evidence: hiring, marketing, and building infrastructure before product-market fit is proven. Startup Genome found 74% of high-growth internet startup failures involve premature scaling, and no prematurely scaled startup in their 3,200-company dataset passed 100,000 users.

How do I know if my startup is failing before it’s obvious?

The decline is usually measurable a year out. CB Insights found 72% of startups that died saw their company health score decline through their final year, dropping 15% on average. Practical leading indicators: flat or declining weekly retention, spend rising without new validating evidence, contribution margin not improving, and founders avoiding hard internal conversations.

Should founders ignore competitors completely?

No. The right cadence is proportional to the actual risk: review competitors monthly, customers weekly. A competitor move matters only if it changes what your customers need from you or invalidates the thesis your spending is built on. Y Combinator’s long-standing advice captures the asymmetry: startups die of suicide, not murder.

Why do funded startups get outcompeted more than bootstrapped ones?

Because venture-scale spending is built on a winning thesis. When a competitor invalidates that thesis, the next round dies and the burn rate does the killing. Bootstrapped companies with positive margins can survive funded rivals indefinitely; they grow slower instead of dying. The competitor removes an assumption; the spend structure delivers the fatal blow.

How do co-founder conflicts kill startups?

Noam Wasserman’s research following roughly 10,000 founders found 65% of high-potential startups fail primarily due to co-founder conflict: disputes over roles, equity, and direction that stall decisions and fracture execution. Founder agreements that pre-decide equity vesting, decision rights, and exit conditions cut that risk by about 44%.

What should I check every week instead of watching competitors?

Five things, in order: weekly retention, one surprising customer learning, whether new evidence justifies the current spend level, whether your core margin number is on a path to positive, and what hard conversation you’re avoiding. Check competitors last, monthly, through one filter: did their move change what your customers need from you?

I write a deep, framework-driven essay like this most days, drawn from two decades of building companies (CommonFloor, Leap.club, and now AI-first ventures). If this one reframed how you think about risk, the sibling essays linked above go deeper on each failure mode.