Knowledge That Becomes Obsolete Before It Is Applied

One of the most durable and reassuring findings in the study of work is the learning curve. Since Wright observed in the 1930s that the cost of building an aircraft fell predictably with cumulative output, and since the experience curve was generalized into strategy decades later, we have organized a great deal of management around a simple faith: the more times an organization does something, the better and cheaper it becomes at it. Knowledge compounds. Practice pays. Experience is capital that accrues interest.

That faith rests on a condition we rarely name, because for most of industrial history it was met automatically. The thing being learned has to stay roughly the same long enough for the learning to be worth the trouble. When you are building the same airframe for a decade, every hour spent getting better at it is an hour well spent. The learning curve is a promise that only holds when the task holds still.

The present moment breaks the condition. The tools, the platforms, the regulations, the customer behaviors, and now the capabilities of the machines we work alongside all change on a timescale that has fallen below the time it takes an organization to learn a lesson, agree on it, write it down, socialize it, and embed it in a procedure. By the time the "lessons learned" review is complete and approved, the conditions it so carefully describes have often already moved on. The organization has become expert, at some cost, in a situation that no longer exists.

It is worth being precise rather than dramatic about this. Not all knowledge depreciates quickly; deep principles age slowly, and much operational know-how remains valuable for years. But in fast-moving domains, a real and growing share of what an institution knows is losing value faster than the institution can put it to use. The half-life of a lesson is shrinking toward, and sometimes below, its time-to-deployment. When that happens, accumulation stops being an unmixed good.

Here is the part our institutions are badly designed for. We have built elaborate machinery for acquiring and storing knowledge, and almost none for retiring it. There is a process to capture a lesson, a repository to hold it, a training program to spread it, and an authority that blesses it as a best practice. There is no equivalent process for declaring that a lesson has expired. So obsolete knowledge does not leave the building. It lingers in the wiki, the playbook, and the onboarding deck with exactly the same authority as knowledge that is still true, and after a while no one can tell the two apart. The archive fills with confident statements that used to be correct.

March's long-running distinction between exploring new possibilities and exploiting known ones takes on a sharper edge in this light. Exploitation—leaning hard on the existing knowledge base—has always been the safer-looking choice. But exploiting a body of knowledge that is quietly rotting is more dangerous than it appears, precisely because it feels like prudence. The organization doubles down on what it "knows," and what it knows is increasingly a photograph of a world that has changed.

The instinctive response—capture faster, document more, shorten the lag between a lesson and its playbook—treats the symptom and feeds the disease. Speeding up acquisition without any corresponding capacity for expiry simply fills the archive faster with material that will age just as quickly, and raises the proportion of the record that is subtly out of date at any given moment. The asymmetry is stark. An organization will invest heavily in the machinery of learning—reviews, repositories, training, certification—and invest nothing whatever in the machinery of unlearning, because no one has ever been promoted for removing a lesson from the canon. We measure, celebrate, and reward how much we know. We do not measure how much of what we know is still true, and in fast domains those two quantities have begun to come apart.

The great management problem of the last century was how to learn faster—how to climb the curve, capture the lesson, spread the practice. That problem is largely solved; we are very good at accumulation. The problem this century may hand us is the opposite one, and we have barely begun to take it seriously: how to tell, quickly and reliably, which of the things we know are no longer true, and how to let go of them before they cost us. Knowing how to learn was the achievement. Knowing what to unlearn may be the frontier.