A Seoul National University study separates two things the solopreneur wave keeps blurring: starting alone and staying.
Ninety-seven percent of them never come back. That is the number sitting inside a dataset of 160,143 Product Hunt launches, assembled by Hyunso Kim, Hyo Kang and Jaeyong Song at Seoul National University's Business School and posted as a preprint in May. Generative AI, they find, has pushed solo founders into the market in numbers nobody has seen before. It has not pushed them to the top. Up there, teams still win, and the gap has widened.
For anyone writing about the one-person company, this is the first piece of evidence that is not a screenshot of somebody's Stripe dashboard. Eight consecutive days of scanning the same niche produced nothing but vendor content and self-reported revenue claims: three million a year, half a million a month, no auditor, no method, no way to check. Here there is a named institution, a dated window, an identification design and a significance level. The reason to read it carefully is not that it confirms the story. It is that it corrects it.
Entry got cheap, endurance did not
The distinction the paper draws is between the cost of entering and the cost of persisting, and these two costs have moved in opposite directions. A model can now do the work that used to require a second and a third person at the start: the landing page, the copy, the first version of the backend, the support responses in the first month. So the founder who would previously have needed cofounders to reach a launch reaches it alone. The entry barrier collapses.
Persisting is a different problem. It requires sales conversations that go badly and get repeated, a decision about pricing that turns out wrong and has to be unwound, a customer who leaves and has to be understood rather than replaced. That work is not bounded by output volume. It is bounded by attention, judgment and the number of hours a person will still be functional after a year. No model has moved that constraint, and the 97 percent one-shot rate is what it looks like when a lot of people discover this at the same time.
That number deserves to be used, not just cited. It is the counterweight to every unverified million-euro-solo-founder claim, and it costs nothing to hold up. The claim may be true. It is also, statistically, a description of the thin tail of a distribution whose fat part is people who launched once and stopped.
The interesting companies are not solo, they are small
Here is where the study sharpens rather than damages the argument about overlooked industries. The thesis was never that lone founders win. It was that people are now running substantial operations in construction, maritime, energy, water and food distribution with teams of one to ten, in sectors where the incumbent structure assumes hundreds. The Seoul paper says teams still lead at the top. That is not a contradiction. It is the same claim with the ambiguity removed.
A two-person company that books eight-figure revenue is not a solo founder story. It is a small-team story that the solopreneur content wave keeps miscounting as its own. The difference matters commercially, because the buyer in a plant or a shipping line is not evaluating whether one person can do everything. They are evaluating whether a team of four with deep domain knowledge and good tooling can deliver what a team of forty used to. The answer to that question is increasingly yes, and it has nothing to do with the 97 percent.
There is a related pattern in the German energy sector that reads like the same finding from a different angle. A PwC study reported this spring found 79 percent of utilities have started AI work, 61 percent cannot get out of pilot, and zero percent have arrived. The stated obstacles were technological in 30 percent of cases, skills shortage in 28, legal and ethical concerns in 18, cultural and organisational resistance in 16. That is 62 percent people and organisation. Entering a pilot is cheap. Finishing is not. Same shape, larger scale.
What the paper does not settle
The honest caveats are real. Product Hunt is a self-selecting population of consumer-facing software launches, which is not the population of businesses being discussed here. A maritime routing operation with four people and three customers never launches on Product Hunt at all. The 97 percent one-shot figure measures repeat launching behaviour, not survival or revenue, and those are different things. And the paper is a preprint, not yet through peer review, which means the identification strategy deserves scrutiny rather than deference.
None of that undoes the direction of the finding. It does mean the number should be quoted for what it measures.
The practical move is to stop asking whether a company can be run by one person and start asking which specific work in a given operation has become cheap enough that four people can now carry what forty carried, and which work is still the judgment a customer is paying a human to exercise.
That question survives the caveats. It also survives the hype cycle, because it does not depend on anybody's revenue screenshot being true. Somewhere in the 97 percent are people who launched once, learned something expensive, and are now doing the second thing quietly with three colleagues and a customer who pays. Those are the companies worth finding. They will not be on Product Hunt.