Why do startups fail, according to the surveys?

CB Insights: a count of public post mortems

CB Insights' report, dated 5 March 2026, is a count, not a survey. CB Insights sells data on private companies. It took 431 venture backed startups that publicly shut down since 2023, none with an earlier exit, and read their post mortems, founder interviews and shutdown announcements. Its analysts chose the categories.

The rest of its chart: wrong market timing or macro conditions 29%, unsustainable unit economics 19%, ineffective strategic pivot 6%, outcompeted by rivals 6%, operational or leadership challenges 5%, technical or clinical issues 3%, fraud or legal problems 3%.

Earlier editions used other samples, and the old address now leads to the new report. The Internet Archive keeps the edition dated 6 November 2019, which read 101 post mortems, and the edition dated 3 August 2021, which read 111 since 2018. The 2021 chart is still on CB Insights' own server.

Money went from 29% to 70% across three editions of one report, on three different samples. In the 2026 one, the median company had raised $11M.

Wilbur Labs: a survey of 200 founders

Wilbur Labs' report, dated 28 April 2026, is a survey. Wilbur Labs is a startup studio, and Wakefield Research helped run it. 200 founders of US tech companies answered online between 3 and 12 February 2026. The stated margin of sampling error is plus or minus 6.9 percentage points.

The same report has external factors at 31%, hiring missteps at 30% and running out of money at 25%, down from 38% in its 2023 survey. Asked for the main lesson of a failure, 54% said understanding product market fit better.

The question joins failing with changing direction, so 45% is not a share of failed companies. And the sample is founders, not companies that closed: 59% said they worry about their business surviving the next 12 months. The method note adds that founders' own accounts may be swayed by attribution bias, exaggeration, telescoping and recency bias.

A peer reviewed study of 50 founders' accounts

A study in Frontiers in Psychology, published 8 February 2024 by six researchers in Hungary, took 50 accounts written by chief executives of failed startups, found by web search in 2018, 47 of them about failures between 2014 and 2017. Three coders marked what each team had lacked, against an existing model of 20 competencies. So it ranks what the team did not do, not reasons such as cash.

Information seeking means gathering data before a business decision. Customer service orientation means putting customers' needs ahead of the technology. The study counted 3.3 gaps per account. The authors name their limit: the chief executives were the only source, which can bias what is recalled and told.

Startup Genome: a benchmark, not a record of closures

Startup Genome's 2011 report on premature scaling rests on about 3,200 high growth internet startups that answered its own survey. An algorithm compared the stage each startup acted as if it had reached with the stage it had really reached, and called a mismatch premature scaling, for example spending too much on customer acquisition before product market fit. About 70% of the sample fell there.

Its headline is that 74% of high growth internet startups fail due to premature scaling. The report does not say how failure was measured, and its caveats say it had no longitudinal data yet. It also says more than 90% of startups fail, with no source.

Where the reports disagree, and why they cannot be added up

• Different companies. Venture backed shutdowns, 200 US tech founders, 50 essays, and startups that answered a benchmark survey. A share from one says nothing about another.
• Several reasons per company. One company sits under several reasons, so the shares pass 100% and are not slices of one whole.
• Different lists of reasons. Each analyst wrote the categories. The top answer is capital in one report, competition in another and information seeking in a third.
• The same essays, counted again. Fractl, a content marketing agency, says many of the 193 failed startups in its 2017 study came from CB Insights' list of post mortems. Skynova, which sells invoicing software, surveyed 492 founders in November 2022, but its method note says its reasons for failure came from CB Insights' post mortems. Reports that look independent share one pile of essays.
• Self reported. Every figure here is what a founder wrote or ticked afterwards. Wilbur Labs and the Frontiers authors both say that can bend the answer.

What no survey can show

• Cause. A post mortem is a story told afterwards by the person who ran the company. CB Insights itself calls running out of capital the final cause of death, not the root problem.
• The survivors. A count of post mortems holds only companies that closed. It cannot say how many companies with the same problem are still trading.
• The silent ones. A public account exists only where a founder wrote one or the closure was reported.
• How many fail. A different question, and the one official statistics answer. Their survival tables count how many new businesses are still trading year by year. They count closures, not reasons. Business survival rates has those figures.

What to do about it

Under different names, all four reports come back to one thing: not enough people wanted the product, and the company found out late.

1. Check that someone needs it. Product market fit says how to tell, and will it sell lists what decides whether a stranger pays.
2. Say who it is for and why it is different: ideal customer and positioning.
3. Run one channel where those people already are. Distribution channels takes each one from its own rules, from Reddit and SEO to Google Ads.
4. Hold the spending until strangers pay. Ads before anyone has paid says when a paid test is worth running.
5. If nobody signs up, or people sign up and never pay, read nobody signing up and signups but no sales before you change channel.