We tend to speak of organizational learning as though it were a cognitive achievement. An organization 'realizes', 'discovers', 'adapts'. The metaphor is seductive and misleading. Organizations do not have minds. What they have are routines, in the sense Barbara Levitt and James March gave the term in 1988: the rules, procedures, technologies, and conventions through which experience is encoded and carried beyond any individual who lived it. Learning, in this view, is the revision of routines in response to experience. Whether that revision occurs depends less on the intelligence of members than on the arrangements that govern which experience gets recorded, who interprets it, and which interpretations are allowed to change anything.

This reframing has a consequence that is easy to state and hard to absorb. Learning is an institutionally structured activity. Institutions are the durable rules, norms, and cognitive schemas that define appropriate action. They determine which signals count as evidence, which actors are credible sources, and which channels carry lessons from one setting to another. An organization embedded in a field governed by professional accreditation learns differently from one embedded in a field governed by venture capital. Neither learns freely.

The analysis proceeds in three movements. First comes the infrastructure that makes learning possible. Second come the pathologies that this same infrastructure produces. Third come the channels through which lessons travel between organizations, carrying the institutional logics of their origins. Throughout, the question is not whether organizations learn. It is what they are structured to learn, and what they are structured to miss.

Every organization possesses a learning infrastructure, whether or not it designed one. The term refers to the arrangements that determine how experience becomes institutional memory: incident reporting systems, after-action reviews, performance evaluations, training curricula, documentation standards, and the less visible conventions governing who may speak in a meeting and what may be said. These arrangements function as filters. They specify in advance which events are noticeable and which disappear into the noise of everyday operation.

Consider the contrast between commercial aviation and many hospital systems in the late twentieth century. Since 1976, the Aviation Safety Reporting System, administered by NASA on behalf of the Federal Aviation Administration, has allowed pilots and controllers to report errors and near-misses confidentially. It also offers limited protection from enforcement action for inadvertent violations. The design is institutional, not technical. By decoupling disclosure from punishment, it changed the incentives surrounding error. It made near-misses, events that would otherwise leave no trace, available as data.

Many hospitals, by contrast, inherited an institutional order that framed error through the logics of professional autonomy and legal liability. Morbidity and mortality conferences existed, but disclosure risked malpractice exposure and reputational damage within steep status hierarchies. The same near-miss that becomes a data point in aviation becomes, in such settings, something to be managed quietly. The difference lies not in the competence or conscientiousness of practitioners. It lies in the architecture surrounding them.

Infrastructure also determines the depth at which learning can occur. Chris Argyris and Donald Schön distinguished single-loop learning, which corrects actions against existing goals, from double-loop learning, which questions the goals and assumptions themselves. Most evaluation routines are built for the former. Key performance indicators measure deviation from a target. They rarely ask whether the target is meaningful. An organization can therefore become highly proficient at correcting errors within a frame while remaining structurally unable to question the frame.

James C. Scott's concept of legibility is instructive here. States, Scott observed, simplify complex realities into forms they can count and administer. Organizational learning systems do the same. They favor knowledge that is codifiable, quantifiable, and attributable. They systematically discount the tacit, practical knowledge, what Scott called mētis, held by those closest to the work. What the infrastructure cannot see, the organization cannot learn.

Takeaway

A learning system is defined less by what it records than by what it renders invisible. Before asking whether an organization learns well, ask what its reporting, evaluation, and memory arrangements make impossible to notice.

If infrastructure enables learning, it also produces characteristic failures. Levitt and March identified several that recur with striking regularity across organizational types. These are not lapses of individual judgment. They are systematic outcomes of how institutions structure the inference from experience to routine. They are often most severe in organizations that believe themselves to be learning well.

The first is superstitious learning: forming confident causal beliefs from ambiguous evidence. Organizations operate in environments where outcomes have many causes, samples are small, and feedback is delayed. When a reorganization coincides with improved results, the reorganization receives credit. When a strategy coincides with failure, it is abandoned. Institutional arrangements amplify the problem. Annual review cycles, executive tenure patterns, and promotion systems reward tidy causal narratives, and those narratives then harden into routine.

The second is the competency trap. As an organization gains experience with a procedure or technology, its performance improves, which increases its use, which improves performance further. This positive feedback lets an inferior but familiar routine outperform a superior but unfamiliar one in the short run, so the superior alternative is never developed. March later generalized this as the tension between exploitation and exploration. Refining existing competence reliably crowds out experimentation, because exploitation yields returns that are nearer, surer, and easier to measure.

The third, and perhaps most consequential, is success-induced blindness. Success generates slack, confidence, and a dominant coalition whose authority rests on the strategies that produced it. Danny Miller described the resulting trajectory as the Icarus paradox: the qualities that elevated a firm become, through elaboration and entrenchment, the causes of its decline. Kodak engineers built an early digital camera prototype in 1975. The difficulty was never ignorance of the technology. It was an institutional order of revenue structures, career ladders, and identity organized around film.

Diane Vaughan's study of the Challenger launch decision reveals the subtlest version of this pathology. Engineers had observed erosion in the solid rocket booster O-rings on previous flights. Because those flights had succeeded, the anomalies were progressively reclassified as acceptable risk. Vaughan called this the normalization of deviance. The organization was learning continuously. What it learned, through entirely routine procedures, was to stop seeing danger.

Takeaway

Learning pathologies are not departures from organizational learning but products of it. The routines that convert experience into confidence are the same routines that convert luck into doctrine and anomaly into normality.

Organizations learn not only from their own experience but from the experience of others. Here the institutional structuring of learning becomes most visible. Paul DiMaggio and Walter Powell's 1983 account of institutional isomorphism identified three mechanisms through which organizations in a common field come to resemble one another: coercive, mimetic, and normative. Each is also a channel of learning, and each shapes what kind of lesson can travel.

Coercive diffusion operates through regulation and resource dependence. After the accounting scandals at Enron and WorldCom, the Sarbanes-Oxley Act of 2002 codified lessons about internal controls and imposed them across publicly traded firms. Regulatory learning has the virtue of reach. It transfers lessons to organizations that would never have learned them independently. Its limitation is that it transmits conclusions rather than the reasoning behind them. This invites compliance that satisfies the letter of the rule while bypassing the understanding.

Mimetic diffusion flourishes under uncertainty. When goals are ambiguous and technologies poorly understood, organizations imitate those they perceive as successful, and consultants serve as the principal carriers. Practices such as Six Sigma, the balanced scorecard, and lean management have spread across sectors far removed from their origins. Consultants abstract a practice from its context, package it as transferable, and sell legitimacy alongside technique. What travels is frequently the visible form of a practice rather than the tacit conditions that made it work where it was first observed.

Normative diffusion runs through professional networks: training programs, associations, conferences, and the movement of personnel. The surgical safety checklist, promoted by the World Health Organization from 2008, illustrates its strengths. Because professionals share socialization and standards of evidence, lessons can travel with their rationale intact. Yet professional channels also filter. They carry what is legible within the profession's paradigm and resist lessons that threaten its jurisdiction or status.

The deepest consequence of institutional diffusion is what John Meyer and Brian Rowan called decoupling: adopting formal structures for their legitimacy while actual practice continues largely unchanged. An organization can import the vocabulary, roles, and documentation of a lesson without learning it. Whole fields can thus appear to converge on best practice while the underlying capacity to perform varies enormously.

Takeaway

Lessons do not travel naked; they travel inside institutional vehicles that determine what arrives. The question to ask of any imported practice is whether you have adopted its substance or merely its legitimacy.

Taken together, these three dimensions dissolve the image of the learning organization as a self-directing intelligence. What an organization can learn is set by an architecture it rarely chose: reporting systems inherited from predecessors, evaluation routines borrowed from its field, and lessons imported through channels that carry their own logics. Learning capacity is less a property of organizations than of the institutional environments in which they sit.

For those responsible for institutional design, the implication is to treat learning infrastructure as a primary object of reform rather than an afterthought. Protecting disclosure, building routines that question goals rather than merely track them, preserving space for exploration, and testing imported practices against local conditions are structural interventions, not cultural exhortations.

The broader lesson is sobering. Organizations rarely fail because they stop learning. They fail because they learn efficiently what their architecture permits, and nothing else.