The Real Startup Failure Rate:
What 43% Actually Means for Your Idea

The 90% figure treats a founder who launched a Shopify store in 2019 and shut it down after three months the same as a founder who raised a $2M seed round, built a 12-person team, and ran for seven years before running out of runway. These are not comparable failure modes. The first is a failed experiment. The second is a preventable loss. Aggregating them into a single statistic does not illuminate anything — it just makes everyone feel equally doomed, which is not useful.

What is useful is the failure data for the specific population you belong to. If you have raised — or are planning to raise — seed capital, the CB Insights post-mortem database is the most relevant evidence available. Here is what it actually says.

What the CB Insights data actually says

The CB Insights 2024 update of its startup failure post-mortem dataset covers 431 VC-backed companies that shut down after raising external capital. These companies collectively represent hundreds of millions of dollars in deployed capital and thousands of person-years of founder effort. The data is self-reported — founders and leadership teams describing in their own words why they failed — which gives it a qualitative texture that a purely quantitative dataset cannot replicate.

The causes are not mutually exclusive — founders cited multiple contributing factors, and a single failure often involved elements of all three. But the ordering matters. Product-market fit failure is number one, and it is the one cause in this list that is detectable before significant capital has been deployed. Running out of cash is often a downstream consequence of misallocating that cash against the wrong market. Hiring the wrong team is partly a downstream consequence of not having a clear enough product thesis to attract the right people. The root often traces back to the same place: an unverified assumption about what the market will pay for, and in what quantities.

The graduation cliff

Carta’s cohort analysis provides a complementary lens. Looking at seed-stage startups that went on to raise a Series A within two years: 30.6% of the 2018 cohort achieved this. For the 2022 cohort, the equivalent figure at the same elapsed time was 15.4%. Source: Carta State of Private Markets, Q4 2024

That is roughly half as many seeds graduating to A, at a median timeline that has extended to approximately 712 days — nearly two years from close of seed to close of Series A. The market has not become more forgiving. It has become considerably less so.

What this means in practice is that every claim in a seed deck now faces materially more scrutiny at the A stage. Investors are better resourced than they were in 2018. They have seen more cohorts fail. They have access to AI-assisted due diligence tooling that can cross-reference market claims against primary sources in minutes. The information asymmetry that once existed between founder and investor — where the founder could simply assert a TAM and have it accepted — has largely collapsed. If your market thesis is not grounded in verifiable evidence, a well-resourced institutional investor will find that out before they sign.

“Product-market fit failure is the cause investors identified most often — and it is the one that is most detectable before you raise. The data says check your claims. Most founders don’t.”

What “product-market fit failure” actually means in practice

“Poor product-market fit” is a category, not a cause. Inside that 43% there are at least four distinct failure modes, each with different diagnostic signatures and different prevention strategies. Treating them as a single thing obscures the point where the problem could have been caught.

The distribution problem

The market exists, and customers want the product — but the founder cannot reach them profitably. Customer acquisition cost exceeds lifetime value at every channel the company has tried. This is a CAC problem disguised as a product problem. The unit economics never worked; the founder assumed distribution would get cheaper as the company scaled. It did not. The diagnostic for this failure mode is simple: stress-test your CAC assumption before you build the acquisition team, not after you have hired it.

The willingness-to-pay problem

The market exists. The pain is real. Customers will use the product enthusiastically if it is free. But when it comes to paying — at any price that supports the unit economics — they do not. Pilots convert to LOIs. LOIs do not convert to purchase orders. Waitlists do not convert to paying customers. The signals that looked like validation during the pre-launch phase were signals of interest, not signals of purchase intent. These are different things, and conflating them is one of the most common mistakes in early-stage validation.

The wrong-solution problem

The market exists, customers would pay for a solution, but not this solution. The product solves a related problem rather than the core one. Customers describe the product as “interesting” but do not integrate it into their workflows. Churn is high. NPS is neutral. The founder interprets this as an execution problem — a feature gap, a UI problem, a positioning problem — and iterates. The underlying issue is that the product thesis was wrong from the start, and no amount of iteration on the execution will fix a structural mismatch between solution and problem.

The market-sizing problem

The market exists but is not the size or at the growth rate the founder claimed. The TAM was constructed top-down from a large market report, divided by an optimistic market-share assumption, without any bottom-up reconciliation. When the company deploys against the actual market, the serviceable addressable market turns out to be a fraction of the headline number. The company runs out of addressable customers before it achieves the scale needed to justify the capital it raised.

All four failure modes share a common characteristic: they are detectable before launch. The signals are available in primary research, in customer discovery, in competitor analysis, and in unit economics modelling. They are routinely missed not because the data is unavailable, but because the validation process did not apply sufficient adversarial pressure to the underlying assumptions.

Why founders miss it

The structural reasons are well-documented, and they are worth naming clearly because awareness alone is not sufficient protection — these are not rational failures that dissolve once you know about them.

Confirmation bias. Founders who have invested months or years in an idea seek evidence that confirms the thesis, not evidence that refutes it. This is not a character flaw — it is a documented feature of human cognition under conditions of high personal investment. The practical implication is that self-directed validation is structurally inadequate: the person asking the questions has a strong incentive to find particular answers, and they will find them.

Proxy metrics. Pilot users, letters of intent, waitlist signups, “strong interest from potential customers,” conversations that “went really well” — none of these are the same as a paying customer in a repeatable acquisition channel. Each of these proxy signals can feel like validation while describing a market that will not actually pay at scale. The discipline required is to ask, at each data point: does this tell me that a stranger would pay for this product in a channel I can operate profitably? If the answer is not clearly yes, the signal is a proxy, not a proof.

Advisor optimism. Most advisors are incentivised to encourage, not discourage. An angel investor who is about to write a cheque has a strong motivation to believe the market thesis. A mentor who has publicly endorsed a founder has a reputational stake in the founder’s success. A friend who provides feedback at dinner is not going to tell you that you have wasted the last six months. The people with the most regular access to a founder are often the least structurally positioned to give honest assessments of the underlying risks.

AI validation loops. As we explored in a companion piece, AI assistants trained on human feedback are structurally optimised to agree with the person using them. Using an AI tool as a validation instrument — rather than a research breadth tool — combines the confirmation bias of self-directed validation with a mechanism that is explicitly trained to confirm whatever direction the conversation has taken. The result is a loop with no adversarial pressure at any point.

What catching it early actually looks like

There is no single test for product-market fit before you have a product. But there is a minimum evidentiary bar that distinguishes a credible market thesis from an unvalidated assumption dressed in the language of confidence. The bar has four elements.

A named failure mode articulated in one sentence — not a list of risks, not a mitigant, not a caveat. One sentence that describes the exact scenario in which this idea does not work. If a founder cannot produce this sentence without hedging, the failure mode has not been confronted. It is still an unexamined assumption.

Unit economics that survive a pessimistic CAC assumption. Not the best-case scenario where the product goes viral and CAC trends toward zero. The scenario where customer acquisition costs 50% more than modelled and conversion rates are 30% lower. If the unit economics break under that assumption, the business model needs rethinking before capital is deployed against it.

A TAM that reconciles top-down and bottom-up. The top-down methodology should be checked against a named primary source — an analyst firm’s methodology document, not a press release. The bottom-up should multiply out from a realistic number of addressable customers, a realistic conversion rate, and a realistic price point. If the two numbers are more than 2x apart, the methodology gap is itself a signal.

A specific named competitor the founder can honestly critique — one whose genuine strengths are acknowledged, not dismissed, and for whom the founder can describe the conditions under which a customer should choose that competitor over you. This test is harder than it sounds. Most founders produce a competitive analysis that describes only their own product’s strengths. An honest competitive critique requires identifying the scenarios where you lose.

A practical pre-raise checklist

Before your deck leaves the room, work through these five questions. If any answer is “no” or “not sure,” that item is the one most likely to stall your process at due diligence.

The 43% is not bad luck

Product-market fit failure is not a random outcome. It is not bad timing. It is not a market that was impossible to predict. In the overwhelming majority of cases documented in the CB Insights post-mortem data, the signals were available before capital was deployed. The assumptions that turned out to be wrong were assumptions that could have been tested. They were not tested, because testing them rigorously required applying adversarial pressure that the available validation tools — advisors, peers, AI assistants — were not structured to provide.

The question is not whether you can achieve certainty before you raise. You cannot. The question is whether you have pushed your assumptions to the point where you can honestly articulate what would have to be true for them to be wrong — and whether you have tested that. The founders in the 43% almost certainly believed they had. The difference between validation and the feeling of validation is the adversarial pressure applied to the assumptions.

Check before you raise. The cost of finding out your thesis is wrong at the pre-seed stage is a changed slide and a week of research. The cost of finding out at month eighteen, after deployment, is everything else.

Not sure where your idea sits against these criteria? The free PRISM probe scores your idea against the same kill lenses in under 3 minutes — no sign-up required. If you want to see how most ideas score in aggregate, our public verdict data publishes every PRISM result monthly.