Is it true that 90% of startups fail? — Evidence and context

Short answer: Does 90% of startups fail? And how many entrepreneurs in the US does that claim involve?

One-sentence verdict

The short verdict is that stating ‘90% of startups fail‘ as a universal fact is misleading because the result depends on what you mean by ‘startup’ and which group of firms you count, and readers looking to assess the claim should consider the number of entrepreneurs in the us who are included in any calculation.

Government survival series track new establishments and show far lower aggregate closure rates than the 90 percent line often attributed to startups, with roughly 20 percent exiting in year one and about half not active after five years according to official data Business Employment Dynamics.

Why the number of entrepreneurs in the US matters for this claim

When someone cites a single percent like 90 you need to ask which entrepreneurs are counted, because the pool might be all new small-business owners, a subset of high-growth founders, or only venture-backed teams; each choice gives a different result and a different policy implication.

For example, counting every new local establishment produces one set of survival numbers, while counting only venture-backed firms or firms in a single tech cohort can yield much higher non-scaling rates in industry studies.

Definitions and context: What we mean by ‘startup’, ‘failure’, and ‘entrepreneur’

Common definitions used by government and industry

Official series such as those from the Bureau of Labor Statistics and the Small Business Administration follow cohorts of new establishments and use the term new establishment to mean a newly recorded business location or firm entry in their administrative data, which differs from industry uses of the word startup that often imply high growth potential or venture backing; see the SBA summary for context SBA Office of Advocacy overview.

Industry reports and investor-focused analyses typically use ‘startup’ to mean enterprises pursuing rapid scale, often measured within venture-backed cohorts or specific accelerator groups, and they may use ‘non-scaling’ or ‘shutdown’ where government data would record an exit or closure.

Why definitions change the percentages you will see

Different definitions change both the numerator and denominator in a survival ratio, so a high non-scaling rate reported for a selective venture cohort does not directly translate to the entire set of new businesses started by entrepreneurs nationwide.

To be precise about claims, readers should distinguish between an entrepreneur who opens a small service business and a founder pursuing rapid national expansion with outside capital; conflating them leads to misleading headline rates.

How government data track survival: What BLS and SBA series actually report

The BLS Business Employment Dynamics series in brief

The BLS Business Employment Dynamics series tracks cohorts of new establishments and reports that about 20 percent of new establishments close within their first year and that roughly half of new establishments are not active five years after entry, according to the BLS cohort tables Business Employment Dynamics.

These figures represent broad aggregation across regions and sectors and include many ordinary small firms such as shops and local services that are not typically described as venture startups.

SBA summaries and the five-year benchmark

The Small Business Administration Office of Advocacy provides accessible summaries of survival patterns and highlights the five-year benchmark as a common reference point for policy and research discussions SBA Office of Advocacy overview.

Government datasets record exits for many reasons including voluntary closure or acquisition, and they do not distinguish reliably between a closure that reflects business failure and an exit that reflects a planned sale or reorganization.

Why industry and venture-backed figures can be much higher: cohort and data differences

What CB Insights and ecosystem reports measure

Industry analyses focused on venture-backed firms or startups aiming for rapid scale report higher non-scaling or shutdown rates because they start with a narrower, higher-risk pool and follow outcomes that matter to investors, and those reports often emphasize different endpoints than government survival series, as noted in university and ecosystem summaries.

Startup Genome and similar ecosystem reports analyze cohorts and regional dynamics and show that selection by funding stage, sector, or geography changes observed outcomes substantially Global Startup Ecosystem Report 2024.

Cohort, sector, and funding stage effects

Venture-backed cohorts exclude many ordinary small businesses and include firms pursuing aggressive growth, so reporting a high percentage of non-scaling firms in these cohorts does not imply the same percentage holds for all new firms created by entrepreneurs across the country.

When industry sources report high failure-like rates for startups they are often describing founder teams that raised external capital or applied to accelerators, and this selection effect alone can account for much of the difference compared with government survival series.

Most common reasons startups stop operating

Top causes identified in post-mortems

Post-mortem analyses of startup shutdowns repeatedly identify lack of product-market fit, cash constraints, and team or operational problems among the most common immediate causes of firms stopping operations, as summarized in industry post-mortems The Top 20 Reasons Startups Fail.

These internal failure modes often interact with external conditions such as a tighter funding environment or weaker market demand, and industry authors emphasize that the mix of causes differs between small owner-operated firms and high-growth teams.

The relative importance of these factors varies by sector, which is why sector-specific reports sometimes report very different non-scaling rates than national averages.

A practical checklist for evaluating a headline statistic like ‘90%’

Questions to ask about source, scope, and definition

Check the original source and identify which cohort is being measured, because a percentage reported for venture-backed firms is not directly comparable to a percentage reported for all new establishments; when in doubt look for the methodology note in the original report Business Employment Dynamics.

Ask whether the author counts acquisitions or voluntary closures as failures, what time window is used, and whether the analysis controls for sector or region, since those choices change headline rates.

How to compare numbers responsibly

Find the primary dataset, read the definitions, and compare like with like: match cohorts by date, funding stage, and sector before concluding that two reported percentages are inconsistent or compatible; the SBA guide shows how survival tables are structured SBA Office of Advocacy overview.

Be cautious of summaries that present a single large percentage without a citation, and prefer tables or appendices that list sample sizes and cohort definitions.

Common errors and misinterpretations when people cite the 90% figure

Conflating cohorts and outcomes

A frequent error is to mix results from selective venture cohorts with broad public series, which inflates perceived failure rates when the cohorts are not comparable; industry and government series measure different populations and outcomes Global Startup Ecosystem Report 2024.

Another mistake is to interpret an acquisition or planned exit recorded in administrative data as a failure, when in many cases an exit is a successful outcome for founders or investors and is counted as an exit in government tables rather than a failure.

Practical examples and scenarios: reading the numbers for small businesses versus venture startups

Example A: A local bakery or small service firm tracked by BLS

Imagine a local bakery that opens in 2022 and is part of a BLS-defined cohort of new establishments; if it closes within a year that exit contributes to the approximately 20 percent first-year closure rate observed in the BLS tables, which aggregate across many such small firms Business Employment Dynamics.

If a bakery remains open but stops employing staff or deregisters over time it contributes to the five-year non-activity rate of about 50 percent reported in government summaries, which is a normal part of business churn and does not imply every closed shop represents a high-profile startup failure.

Example B: A venture-funded app startup analyzed by CB Insights

Now imagine a small team that raises venture funding to build a national app; industry post-mortems may classify many such startups as non-scaling if they cannot reach sufficient revenue or user growth, and reports focused on these cohorts commonly show higher shutdown or non-scaling rates than general establishment series The Top 20 Reasons Startups Fail.

Because venture cohorts are a selective sample of the broader set of new firm entries, their reported non-scaling percentages should not be applied to the entire population of entrepreneurs in the US without clear justification.

Evidence-based takeaways for readers, voters, and local stakeholders

Summing up what the data reliably show

According to government survival series, roughly 20 percent of new establishments exit in year one and about 50 percent are not active after five years, which documents substantial early attrition for new businesses but falls well short of a universal 90 percent claim SBA Office of Advocacy overview.

Industry reports indicate that many venture-backed or high-growth cohorts experience higher non-scaling rates, but these figures vary by cohort, sector, and funding stage and do not translate directly into a single national percentage for all entrepreneurs.

What is still uncertain and open research questions

Open questions for policymakers and researchers include creating standardized longitudinal public datasets that separate venture-backed startups from ordinary small-business entries and clarifying definitions of failure versus exit so that cross-study comparisons become more reliable, a need highlighted in ecosystem reviews and working papers working papers.

Where to find the primary sources and how to keep the number in context

Direct links to BLS, SBA, CB Insights and ecosystem reports

Primary sources to consult include the BLS Business Employment Dynamics site for cohort data, the SBA Office of Advocacy survival summaries, the Startup Genome ecosystem reports, CB Insights post-mortem analyses, OECD business-demography indicators, and working papers that review exit and survival evidence Global Startup Ecosystem Report 2024.

When verifying a headline percentage check the cohort definition, the time window used, and whether exits include acquisitions or planned sales; these three checks will reveal why two numbers that look similar may mean very different things.

Quick tips for verifying statistics in future reporting

Three quick verification steps are: identify the cohort, locate the original table or methodology note, and compare the time windows and outcome definitions; these steps reduce the chance of repeating a misleading headline.

Maintaining clear distinctions between small-business churn and venture cohort outcomes helps voters, local stakeholders, and journalists interpret what the numbers imply for local economic opportunity and for questions about support for entrepreneurs.