The 90 percent myth and what survivors do differently

Why do startups fail?

Startups fail because demand never arrives at the price and the pace the plan assumed, and the cash runs out before the team finds out why. Ask why do startups fail and the ranked evidence gives four overlapping answers: capital exhaustion is cited in 70 percent of venture-backed shutdowns since 2023, poor product-market fit in 43 percent, bad timing or macro conditions in 29 percent, and unsustainable unit economics in 19 percent¹. Only the first of those is visible on the day the company closes.

The shares sum well past 100 because a single shutdown usually names two or three factors together. That overlap is the useful part. Companies rarely die of one clean cause. They die of a chain that starts with a weak demand signal and ends at an empty account.

The ranking comes from an analysis of 431 venture-backed companies that shut down since 2023¹. It is the sharpest current dataset on startup failure reasons, and it describes the funded end of the market rather than the whole economy. For the wider picture, the survival numbers below are the better guide.

What is the real startup failure rate?

Lower than the number everyone quotes. The claim that 90 percent of startups fail has been repeated since at least 2015 with no primary research behind it, and no government or academic dataset supports it as a general figure. US Bureau of Labor Statistics data shows about 48.6 percent of new employer establishments close within five years, meaning 51.4 percent are still trading, and that band has stayed between 45.4 and 51.9 percent since 1994².

Definitions carry most of the argument. The startup failure rate is the share of new ventures that cease independent operation, whether they shut down, are sold for parts, or return no capital, measured over a stated window, most often year one, year five and year ten. Change the window or the population and the answer moves by tens of points. That elasticity is exactly how the 90 percent figure survives fact-checking: it is never attached to a window or a population, so it can never be wrong.

Read as startup survival statistics rather than failure statistics, the curve is steady rather than dramatic. Roughly 80 percent of new establishments make it past year one, 68 percent past year two, 51 percent past year five and 33 percent past year ten². Attrition is heaviest early, then slows.

Sector changes the odds sharply. About 63 percent of tech businesses fail within five years, and 75 percent of venture-backed fintech startups fail⁴. Venture funding raises the ceiling and the failure rate at the same time, because the model is underwritten against outcomes that almost never land.

Is running out of money the reason, or just the trigger?

The trigger, in almost every case. Capital exhaustion is the last event in the chain, which is why it tops the list and why the list misleads if you read it as a ranking of root causes.

Ran out of cash is the death certificate. It is rarely the cause of death.

The timing supports that reading. The median venture-backed company that failed had raised roughly 11 million dollars and closed 22 months after its last round¹. Close to two years passed between the money landing and the doors closing. Whatever broke had a long window in which it was visible, and in most cases fixable.

The pattern holds outside venture too. Most small businesses that fail name cash flow as a cause, which is less a diagnosis than a description of the final quarter of trading.

Reading the ranked causes against each other suggests a consistent upstream story. The mapping below is our inference from those rankings, not a measured finding.

What does product-market fit look like before it disappears?

Product-market fit is the point at which a product meets strong, sustained demand: organic growth, high retention, and customers pulling the product rather than the company pushing it. Its absence is the most durable finding in the failure literature. An earlier CB Insights post-mortem analysis put no market need at 42 percent of failures⁶, close to the 43 percent that cite poor product-market fit in the current data¹. The label changed. The problem did not.

The practical failure is measurement. Teams read top-of-funnel growth as fit, when fit lives in the second and third month of a cohort. Three signals separate real fit from a paid impression of it.

• Retention flattens instead of decaying. A cohort curve that levels off has found a job to do. One that keeps sliding has not, at any acquisition volume.
• The buyer names the problem before you do. If every discovery call requires you to explain why the problem matters, you are selling education, not a product.
• Usage concentrates. Fit shows up as a narrow set of customers using a narrow set of features constantly, not as broad shallow usage spread across segments.

Survivors respond to a weak signal by narrowing. They cut the customer definition until the retention curve reads clean, then expand from a position they can defend. Chasing total addressable market early does the opposite: it adds segments that each behave slightly differently and makes the aggregate unreadable. A structured design thinking process earns its keep here mainly because it forces the problem definition to be written down before the build starts, which turns a later pivot into a decision rather than an argument.

Why do startups fail on timing rather than product?

Because a team can build the right thing 18 to 24 months before the market or the capital environment is ready, watch growth stay flat, and conclude they built the wrong thing. Bad timing or macro conditions are cited in 29 percent of recent shutdowns¹, and timing failures are the hardest to read from inside the company, because they produce the same dashboard as a product failure.

The distinguishing test is qualitative, not quantitative. In a product failure, prospects understand the pitch and do not want it. In a timing failure, prospects want it, agree on the value, and cannot find budget, procurement approval or an internal owner this year. The second group is worth staying alive for. The first is not. Founders who cannot separate the two either abandon a good position early or fund a hopeless one for years.

What do the surviving startups do differently?

They treat runway as a forcing function rather than a countdown. Runway is the number of months a company can operate at its current burn before the cash is gone, and the decision it should drive is whether to cut, raise or pivot. In survivors, a growth stall triggers a cut in burn and a re-test of the core value proposition. In failures, it triggers a bridge round that buys more time on the same plan.

Four habits recur.

1. Re-underwrite the thesis every two quarters. Write down what would have to be true for the current plan to work, then check it against cohort data instead of against the pitch deck.
2. Cut before the runway forces it. A reduction made with twelve months of cash is a strategy. The same reduction at four months is a fire sale of the team’s confidence.
3. Narrow under pressure. When the signal is ambiguous, survivors remove segments. Failures add them, then mistake the resulting noise for a market.
4. Fix the founding team problem early. Co-founder conflict and weak founder-market fit turn up repeatedly in postmortems and almost never on the official cause list, because companies that die that way are recorded as having run out of money.

None of that requires more capital. It requires a build process that can absorb a change of direction without restarting, which is the practical case for treating product development as a discipline with checkpoints rather than one long push to launch. Teams without the internal bandwidth to run those checkpoints often bring in product development consulting for the discovery and validation stages specifically, where the cost of being wrong is lowest.

What warning signs show up first?

Failure is legible well before it is fatal. The signals that matter are visible in a normal operating month, and none of them require a data team to spot.

• Retention curves that never flatten, at any channel or price point.
• Revenue growth that tracks sales headcount almost exactly, which means the product is not compounding.
• A roadmap driven by the loudest customer rather than the largest repeated pattern.
• Payback periods that only work at discounted acquisition cost.
• A founding team that has stopped arguing about strategy, which usually means someone has quietly disengaged.

Any one of these is survivable. Three at once, with under twelve months of runway, is the point at which the honest options are a cut, a narrowing or a pivot, taken in that order and taken now.

Where to start

Pick the single assumption the plan depends on most and test it against cohort data this month rather than this year. Most companies that end up on the failure list had close to two years between their last raise and their close¹, and spent it defending the original plan instead of interrogating it. If a second read on where a product sits against this evidence would help, talk it through with our team.

Written by the group's editorial team with the practice leads who run these builds. Reviewed before publish. Spotted an error? Tell us and we will fix it.