ISSUE 001
Startup Development
I went looking for solid numbers on why startups fail and mostly found the same statistic copied between about forty websites, each one citing the last. So this entry ended up being partly about the statistic and partly about how it got everywhere. The short version: the number is probably directionally right, its evidence is weaker than its fame suggests, and the fix it points to costs nothing but nerve.
An idea worth thinking about
Founders writing their own obituaries is the weakest evidence we have, and it's most of what we have
The startup failure canon is built almost entirely on post-mortems — essays written by founders and investors after the company died. CB Insights has been collecting them for over a decade and now has 483 in one compilation. It's genuinely the best assembled record of startup death anywhere, and it's worth reading. It's also self-reported, retrospective, and written by people with reputations to manage, which is close to the definition of unreliable testimony.
Consider what that does to the answers. A founder who ignored six months of lukewarm customer conversations is unlikely to write "I didn't listen." Far easier to write that funding dried up, or the market wasn't ready, or a competitor moved faster. Every one of those can be true and still be a story told from the most flattering available angle. The failures nobody writes up at all — the quiet ones that never raised money — are missing from the record entirely.
I'm not saying discard it. I'm saying notice that the most repeated claim in startup advice rests on the softest available data, and treat it as a hypothesis you can test cheaply on your own idea rather than a law of nature.
Read the post-mortem collection
A tool I disappeared into
Three decades of dead startups in one spreadsheet
Somebody took the CB Insights post-mortem compilation and turned it into a downloadable dataset: 483 failed companies, each with its sector and its founding and shutdown years, spanning 1992 to 2024. Free on Kaggle. It is not analysis, just the raw list, which is what makes it interesting — you can sort by lifespan and see how long these companies actually lasted before folding, or filter to a sector and read the years like a weather pattern.
Two limits worth stating. It inherits every bias in the underlying collection, so it's weighted toward notable venture-backed companies and tells you nothing about the ordinary business that closed without an essay. And it's a snapshot ending in 2024, not a live feed. Useful for pattern, useless for currency.
Open the dataset
Learn something
Y Combinator gives its whole curriculum away, no application
Startup School is YC's free online course, and free means free — no application, no equity, no cost. It runs about seven weeks at one to two hours a week, self-paced, and the lectures are taught by the partners themselves: Michael Seibel on planning an MVP, Carolynn Levy on modern startup funding, Gustaf Alstromer on getting users. You also get the co-founder matching platform, which has made over 100,000 matches, and a weekly update tool for tracking your own progress.
The honest caveat is that the advice is calibrated for venture-scale software companies aiming at a YC application, so some of it doesn't transfer to a business that intends to stay profitable and privately held. The customer-conversation material transfers to almost anything.
Sign up for Startup School
What we’re seeing
Running out of money shows up in most failures. It's almost never the actual cause.
Here's the part of the failure data I think is genuinely useful, and it's a distinction rather than a number. Cash exhaustion appears in the large majority of startup shutdowns — one analysis of the 2024 CB Insights update puts it in roughly 70 percent of cases. But it is the last thing that happens, not the first. The money runs out while the company is chasing demand that was never confirmed to exist.
The reframing matters because it changes what you'd do about it. If cash is the cause, the answer is raise more, and plenty of founders have raised more and failed anyway. If cash is the symptom, the answer is find out whether anyone wants this before you spend eighteen months building it — which is cheaper, faster, and considerably less pleasant, because it involves being told no by real people early.
A note on provenance, since this entry is partly about provenance. CB Insights' own compilation of 483 post-mortems is verifiable, with its most recent installment covering October 2023 through May 2024. The tighter version of the claim — that the 2024 update analyzed 431 venture-backed shutdowns and recast the top cause as poor product-market fit rather than no market need — comes from secondary analyses of that update rather than from CB Insights directly. Directionally consistent across several independent readings, but not something I'd put a decimal point on.
What survives all the uncertainty is unglamorous and hard to argue with: the most common way to kill a company is to build first and ask second.
CB Insights, 483 startup failure post-mortems
One question
How many people did you talk to before you started building?
A number, not a range. Reply with it — I'm curious where the honest distribution sits, and I'll report the spread once enough of you have answered.
More in this pillar as it's written.
— Shana
Sources link to the original publisher, never to a summary.