A ninety percent failure rate belongs in research, not payroll

Alan Kay’s working life gives the number some meaning. At Xerox’s Palo Alto Research Center in the 1970s he led the group that built Smalltalk, a language whose ideas about objects shaped most programming since, and worked on the overlapping-window interfaces that later reached the public through the Macintosh. He also described, in 1972, a portable computer for children he called the Dynabook, a flat device with a screen and keyboard that nobody could build with the parts of the time. Much of what PARC attempted went nowhere, and much of what worked was commercialized by other companies. A few results changed the industry.

The line is widely credited to Kay, though the occasion on which he first said it is hard to pin down. The idea is the arithmetic of research. If each ambitious project has a small chance of success, a researcher who succeeds most of the time is probably choosing projects that were never very ambitious. A lab that aims high and hits one time in ten can produce more value than one that aims low and always hits, because the hits are so much larger.

The same arithmetic runs venture capital. Most investments in a fund lose money, and one or two return the rest several times over. Investors who want no failures end up with no outsized wins either.

The trouble comes when the line is carried into places where the arithmetic does not hold. A single startup is not a portfolio. It has one bank account, and a founder who fails at ninety percent of what matters to survival, such as payroll, the main product or the key customer, will not be around for the tenth attempt. The advice fits the experiments a company runs, such as new features, marketing channels or prices, where many small failures are cheap. It does not fit the core.

Some work should aim for almost no failures at all. Accounting, security, medical devices and anything customers rely on daily reward reliability, and a team there that fails often is simply doing its job badly.

The number itself is rhetorical. Nobody measures their failure rate to the percent, and a team that chases a high one can do so by picking impossible goals, which looks like ambition and teaches nothing. The better reading is that the ambitious parts of the work should be allowed to fail often, that those failures should be cheap and fast, and that the rest of the business should be run so that it rarely fails at all.