Institutions often think about bottlenecks as problems of insufficient resources. A process moves too slowly because there are not enough staff, a digital system lacks capacity, an approval procedure contains too many stages or information takes too long to travel between organisational units. Yet some of the most consequential bottlenecks in an institution can be much less visible because they are also among its greatest strengths. A highly experienced official understands a policy domain better than almost anyone else, a specialist can interpret evidence that colleagues struggle to evaluate, a manager knows how to resolve unusual cases, or a technical expert has become indispensable whenever a complex decision must be made. From the perspective of local capability, these people are extraordinarily valuable. From the perspective of the wider system, however, the concentration of knowledge, judgement or interpretive authority around them can create a dependency through which increasing amounts of institutional work must pass. Exceptional local capability can therefore coexist with global fragility, because the person who enables the institution to solve difficult problems may also become the point at which its cognitive throughput is constrained.

This is not a contradiction so much as a consequence of specialisation. Complex institutions inevitably depend on people who know more about particular problems than their colleagues, and there would be little value in trying to distribute every form of expertise evenly across an organisation. Expertise takes time to develop, some responsibilities require unusual combinations of experience and judgement, and certain problems arise too infrequently to justify creating large numbers of specialists. Concentration can therefore be efficient and entirely rational. The difficulty begins when the institution becomes unable to perform a recurring cognitive function without routing that function through one particular person or a very small group. At that point, expertise concentration has become systemic dependency, and the institution’s capacity to process work is no longer determined simply by how capable its many components are, but by how much demand a critical individual can absorb.

The pattern often develops gradually. A capable employee becomes the person colleagues consult when something unusual occurs. Because that person gives reliable answers, more difficult cases are referred to them, allowing their expertise to deepen further while other employees gain fewer opportunities to develop comparable judgement. Managers begin to rely on the arrangement because it works, and procedures evolve around the assumption that exceptional cases can always be escalated to the expert. Over time, the person becomes increasingly valuable and increasingly difficult to bypass. Nothing in this process requires poor management or deliberate knowledge hoarding; indeed, it can emerge precisely because everyone is behaving sensibly. Yet the success of the local solution can progressively create a system in which one individual’s capability becomes part of the infrastructure on which everybody else’s work depends.

This dynamic reveals why CAPACITY ≠ THROUGHPUT. An institution may contain a great deal of expertise in aggregate while remaining unable to process certain decisions at the rate required by its environment. If hundreds of employees can complete most of a process but a critical subset of cases ultimately requires interpretation by one specialist, adding more people elsewhere may have almost no effect on overall responsiveness. Work simply accumulates before the scarce cognitive function. The bottleneck is not necessarily the amount of labour available to the organisation, but the concentration of a particular form of judgement that cannot easily be substituted, distributed or reproduced. In such circumstances, the institution can appear well resourced while remaining structurally constrained by a very small part of its cognitive architecture.

The same problem can occur with authority as well as expertise. Some people become bottlenecks because too many decisions require their approval, even when others possess enough knowledge to make those decisions competently. Senior officials may gradually accumulate exceptions because the organisation responds to uncertainty by escalating upwards, or managers may retain decision rights because accountability structures make delegation difficult. Here the scarce resource is not necessarily superior knowledge but the institutional permission to make judgement consequential. A capable senior official can process these decisions quickly and intelligently, yet the system remains dependent on that person’s availability. Meetings, travel, competing priorities or illness can then slow activities across multiple teams, revealing how much institutional throughput had become attached to one node.

Interpretive bottlenecks can be even harder to see. Organisations frequently contain people who act as translators between professional worlds: someone who understands both policy and technology, both legal constraints and operational practice, or both scientific evidence and administrative decision-making. These people make coordination possible because they can translate categories, recognise misunderstandings and explain why information generated in one domain matters in another. Their contribution may be difficult to capture in a job description, but colleagues gradually learn that complicated cross-boundary problems need to pass through them. The institution becomes cognitively more capable because the translator exists, while simultaneously becoming more fragile because the relationship between several parts of the organisation depends upon a capacity concentrated in one person.

This is why key-person dependency should not be understood only as a succession risk. The familiar question — what happens if this person leaves? — is important, but it captures only one part of the problem. A person can remain in post for many years and still function as a bottleneck because the volume of institutional demand exceeds the amount of attention, judgement or translation that one human being can provide. The relevant risk is therefore not merely disappearance but saturation. As more work becomes dependent on a trusted expert, the organisation may respond by asking that person to work harder, prioritise more aggressively or delegate simpler tasks, but none of these responses necessarily changes the underlying architecture. The institution continues to possess a scarce cognitive function whose throughput is bounded by individual capacity.

Technology can sometimes reduce this dependency, but only when the bottleneck has been understood correctly. If a specialist spends much of their time retrieving information, performing repetitive analysis or applying stable decision rules, digital systems and AI may allow part of that work to be distributed or automated. Yet if the person’s value lies in interpreting ambiguous cases, connecting knowledge from several domains or recognising when established rules no longer fit the situation, automation may reproduce only the visible surface of their work. A poorly designed technological intervention can even strengthen the bottleneck if new systems generate more exceptions that still require the same expert to resolve them. The question is therefore not whether technology can replace the person, but which cognitive function has become scarce and whether that function can be supported, distributed or reproduced without destroying the judgement that made it valuable.

The same caution applies to documentation and training. Writing down what an expert knows can improve institutional resilience, but not all expertise can be converted into explicit instructions. Experienced practitioners often recognise patterns before they can fully articulate why those patterns matter, and their judgement may depend on contextual distinctions accumulated over years. Treating documentation as a complete solution can therefore create the illusion that dependency has been removed when the difficult interpretive work remains concentrated. A more robust response may require changing how others participate in difficult cases, creating opportunities for judgement to develop through practice, redesigning escalation pathways or distinguishing between knowledge that can be standardised and knowledge that requires sustained professional development.

Importantly, reducing dependency does not mean reducing the expert’s value. Institutions sometimes respond to bottlenecks by trying to make exceptional individuals less central, as though the problem were their influence rather than the architecture surrounding it. This can destroy valuable capability without creating anything to replace it. The better objective is to prevent the institution from requiring a scarce individual to perform every instance of the function they perform best. An expert can become more valuable when routine demand no longer consumes their attention and their knowledge is used instead to strengthen the wider system, develop others, improve decision conditions and concentrate on genuinely exceptional problems.

The distinction between local capability and system capability becomes especially important when institutions evaluate performance. A team containing an outstanding individual may achieve excellent results, leading managers to conclude that the organisational arrangement is working well. Yet high performance can conceal dependency if the result would deteriorate sharply without that person’s continuous intervention. Conversely, a system that deliberately distributes capability may initially appear less efficient because more people are learning, judgement is being made in several places and some redundancy has been introduced. From a narrow perspective, this can look like duplication; from a systemic perspective, it may represent increased resilience and throughput because the organisation is no longer relying on a single cognitive pathway.

This does not imply that every capability should be duplicated. Some expertise is genuinely rare, and maintaining several copies of every specialist function would be economically impossible. Robust institutional design therefore involves deciding where concentration is acceptable and where dependency becomes dangerous. A person who handles a handful of genuinely exceptional cases each year may represent sensible specialisation; a person through whom hundreds of ordinary decisions must pass may indicate an architectural problem. The relevant issue is not concentration itself but whether the wider institution can continue to function intelligently when demand increases, circumstances change or the individual becomes temporarily unavailable.

Seeing people as possible bottlenecks therefore requires a change in perspective. The question is not who is slowing the organisation down, which would turn a systemic diagnosis into an accusation against individuals, but which institutional functions have become dependent on the throughput of particular people. That formulation makes it possible to recognise that the bottleneck may exist precisely because the person involved is exceptionally capable. Their competence attracts demand, their reliability increases dependence and their success can delay recognition that the surrounding system has become fragile. What appears to be an individual strength can consequently reveal an institutional design problem when too much collective cognition depends on that strength remaining continuously available.

Institutions will always need experts, trusted decision-makers and people whose unusual judgement allows difficult problems to be solved. The goal is not to eliminate these concentrations of capability but to understand their systemic consequences. A robust institution is not one in which nobody is indispensable because everyone possesses identical knowledge; it is one that can benefit from exceptional individual capability without allowing the throughput of the whole system to become hostage to a single person’s attention, judgement or availability.