We dissected 21 famous failures — real companies, real post-mortems, real sources. Then we fed each one to Kasspian with the name stripped off, before it knew how the story ended. Average score: 2.6/10. Here’s what actually kills startups — and the one fast test that would have caught each.
Cause of death across the 21 dissected, most common first.
The pattern is consistent: most startups don’t die in a fair fight with a competitor. They die building something the market never wanted — the one thing the fastest test would have told you first. How we ran this, and what it doesn’t prove →
Juicero
Consumer hardware
A company raised $120 million to sell a juicer you didn't need.
Cause of death · No market need
The test that would’ve caught it: That customers will stay subscribed to the produce packs long enough — six to twelve months — for the business to recover the cost of the machine it gave them.
Read the autopsyQuibi
Media & streaming
Quibi raised $1.75 billion and lasted six months.
Cause of death · No market need
The test that would’ve caught it: That people will pay a separate monthly subscription specifically for 10-minute premium episodes watched only on a phone — when the short-form habit, and the supply, already lives for free on TikTok and YouTube.
Read the autopsyMoviePass
Subscription
MoviePass charged $9.95 a month and paid the theater full price every time you went.
Cause of death · Flawed business model
The test that would’ve caught it: That selling something below cost to grow fast will convert into a profitable business — via data, ads, and studio deals — before the cash runs out.
Read the autopsyWebvan
E-commerce
Webvan built warehouses for a hundred cities before proving it could win one.
Cause of death · Scaled too early
The test that would’ve caught it: That razor-thin grocery margins and costly home delivery can support hundreds of millions in custom warehouse infrastructure — built nationwide up front, before a single city has shown the orders and the economics actually work.
Read the autopsyTheranos
Healthtech
A $9 billion company was built on a machine that never worked.
Cause of death · Fraud / regulation
The test that would’ve caught it: That the core technology — accurate results from a single drop of blood across hundreds of tests — actually works, and can be independently proven before a single real patient relies on it.
Read the autopsyRabbit R1
Consumer hardware
Rabbit sold a hundred thousand people a $199 AI box to run their apps, then most of them stopped using it within weeks.
Cause of death · No market need
The test that would’ve caught it: That a small standalone gadget can reliably drive the apps and services on your phone better than your phone can — and that enough people will carry and pay for a second device to do, less reliably, what the one already in their pocket does for free.
Read the autopsyNate
E-commerce
The AI doing your shopping was hundreds of people in a Manila call center, clicking buy by hand.
Cause of death · Fraud / regulation
The test that would’ve caught it: That a single AI agent could reliably complete checkout on any merchant on the internet — thousands of unique, constantly changing, anti-bot-defended flows — with no integrations and no human stepping in, and that this was already solved at 90-plus percent.
Read the autopsyTerra / LUNA (Terraform Labs)
Crypto
They called it a stablecoin. It was a confidence game with a yield bolted on.
Cause of death · Flawed business model
The test that would’ve caught it: That a $1 peg can be held by arbitrage against a volatile sister token — and that a fixed 19.5% yield paid out of a finite reserve is a growth engine rather than a countdown timer.
Read the autopsyFTX (Sam Bankman-Fried)
Crypto
They backed billions in customer money with a coin they printed themselves.
Cause of death · Fraud / regulation
The test that would’ve caught it: That a token the company prints itself can be counted as real collateral backing real customer deposits.
Read the autopsyAmazon Fire Phone
Consumer hardware
Amazon built a phone around a feature nobody asked for.
Cause of death · Outcompeted / no moat
The test that would’ve caught it: That the standout feature — point your camera at an object to buy it on Amazon — will work reliably in the real world and save people enough effort to be a reason to switch phones.
Read the autopsyGoogle Glass
Wearables
Google built a face computer before anyone agreed it was okay to wear one.
Cause of death · No market need
The test that would’ve caught it: That people will wear an always-on camera on their face in public — and that the people around them will accept being near it — before any of the features matter.
Read the autopsySegway
Mobility
It was hyped to reshape cities. It sold to mall cops and tourists.
Cause of death · No market need
The test that would’ve caught it: That a $5,000 personal transporter will change how ordinary people make short trips — when adoption actually depends on price, on where you're legally allowed to ride it, and on a habit that may never form.
Read the autopsyHomejoy
Marketplace
Homejoy matched cleaners with customers — then watched them swap numbers and cut it out.
Cause of death · Flawed business model
The test that would’ve caught it: That customers and cleaners will keep booking through the app — and keep paying the platform's cut — rather than simply exchanging numbers after the first job and arranging everything directly from then on.
Read the autopsyPets.com
E-commerce
Pets.com went from IPO to liquidation in 268 days.
Cause of death · Scaled too early
The test that would’ve caught it: That you can sell heavy, low-margin commodity goods below cost, subsidise the shipping, and still reach profitability before the cash spent acquiring each customer runs out.
Read the autopsyJawbone
Wearables
Jawbone raised $900 million and still lost the wristband.
Cause of death · Ran out of cash
The test that would’ve caught it: That a hardware startup can hold a defensible position in a category that giants with cheaper supply chains, app platforms, and phones already in everyone's pocket will commoditise.
Read the autopsyOuya
Gaming
Ouya sold the console and forgot it needed the games.
Cause of death · Outcompeted / no moat
The test that would’ve caught it: That players will buy a new console before there are great games for it, and developers will build great games before there are players — the chicken-and-egg both sides of a platform face at once.
Read the autopsyBetter Place
Cleantech
Better Place spent $800 million building stations before it had drivers.
Cause of death · Scaled too early
The test that would’ve caught it: That carmakers will standardise their vehicles around your swappable battery, and that you can pre-build a capital-intensive national network before enough drivers exist to pay for it.
Read the autopsyJibo
Robotics
Jibo was a $900 robot doing a $40 speaker's job.
Cause of death · No market need
The test that would’ve caught it: That households will pay several hundred dollars for charm and personality, when the useful jobs the robot does are about to be done by a smart speaker that costs a tenth as much.
Read the autopsyShyp
On-demand
Shyp charged $5 to do something that cost it far more.
Cause of death · Flawed business model
The test that would’ve caught it: That a flat $5 fee can cover a human courier travelling to you, packaging an item, and handling the shipment — and still leave a margin once you're outside a dense city.
Read the autopsyMunchery
Food delivery
Munchery cooked more dinners than it could sell — every single day.
Cause of death · Flawed business model
The test that would’ve caught it: That you can forecast daily demand closely enough to cook fresh meals in your own kitchens without the unsold food — and the thin margins on prepared meals — quietly eating the business.
Read the autopsyHumane AI Pin
Consumer hardware
Humane built a $699 device to replace the smartphone before it proved anyone wanted to put their phone down.
Cause of death · No market need
The test that would’ve caught it: That a meaningful number of people want a standalone, screenless AI device to replace the phone they already own and love — and will pay around $700 plus a subscription for a slower, less capable version of things their phone already does instantly.
Read the autopsyEach company was rewritten as a one-line pitch with the name and any give-away detail stripped out, then put through the same free scorer anyone can use — no special access, no thumb on the scale. One run each, cold. The exact pitch we fed it is printed at the top of every autopsy, so you can check our work, or paste it in yourself and see what you get.
The honest limit: these are famous failures, and we can’t prove the model didn’t recognise them.Stripping the name doesn’t strip the training data — a pitch about a $699 screenless AI pin is still recognisably Humane. So this is not a prediction test, and we won’t dress it up as one. It can’t tell you the scorer would have called Webvan in 1999. What it shows is whether the tool independently lands on the assumption the post-mortem later proved fatal. A consistency check, not a crystal ball.
Two more things worth saying out loud. One cause each is a simplification — real companies die of several things at once, and FTX was a flawed model and a fraud. We tagged the dominant mechanism from the published post-mortem, nothing cleverer. And 21 famous flameouts is not a random sample of startups: these are the deaths big enough to get written about. Read the 33% as the pattern across these 21, not as a statistic about startups in general. At this sample size, reclassifying a single company moves a bucket by about five points.
What is the most common reason startups fail?
The leading cause is no market need — 33% of the 21 famous failed startups we dissected died from it.
What are the main reasons startups fail?
Across the 21 dissected, the top failure modes are no market need (33%), flawed business model (24%), scaled too early (14%).
Can startup failure be avoided?
Most startups don't die fighting a competitor — they die building something the market never wanted. The fastest, cheapest test would have caught it before the year was burned.
Does the AI just recognise these famous companies?
We can't rule it out, and we don't claim otherwise. The name is stripped from every pitch before it's scored, but stripping a name doesn't strip the training data — a description of a $699 screenless AI pin is still recognisably Humane. So this isn't a prediction test. It's a consistency check: does the tool independently flag the assumption the real post-mortem later proved fatal? Judge it on that, not on foresight.
Every startup in this graveyard had one fatal assumption. Kasspian finds yours — then shows you the fastest way to test it before you build.
Pressure-test my ideaOne week, a dead startup and why it died. The next, a play to get your next customers.
No fluff, no spam. Unsubscribe anytime.