When an institution wants greater consistency, control or strategic coherence, centralising decisions can seem like an obvious solution. Authority that was previously dispersed across local offices, departments or professional units is moved towards a smaller number of decision points, allowing the organisation to establish common priorities and reduce contradictory action. In administrative terms, the result may genuinely be more centralised. Yet something important does not necessarily move with the authority: the knowledge required to make those decisions well.
Knowledge has a geography of its own. Some of it can travel relatively easily because it is codified in records, indicators, reports or databases. Other forms remain closely connected to particular environments, professional practices, relationships or accumulated experience. A local official may understand why a formally straightforward intervention repeatedly fails in one community. A frontline professional may recognise an unusual case that disappears inside an aggregate category. A regional office may know that the apparent meaning of a national indicator changes under particular territorial conditions. Moving the authority to decide does not automatically move these forms of knowledge to the same place.
This creates a paradox. CENTRALISING AUTHORITY ≠ CENTRALISING KNOWLEDGE. Indeed, centralising decision authority can sometimes make the knowledge relevant to those decisions more distributed relative to the place where judgement now occurs. The people making the decision may possess greater formal authority while becoming more cognitively distant from the circumstances to which that authority is applied.
The problem is not simply that central decision-makers lack information. Information can often be transmitted. Local offices can submit reports, administrative systems can provide real-time data and central teams can consult specialists across the organisation. The deeper question is whether the knowledge needed to interpret those inputs travels with them. A figure that is meaningful to someone embedded in a particular operating environment may be ambiguous to someone observing it from the centre. A local exception may look like inconsistency when its contextual rationale is invisible. A pattern that appears obvious in an aggregate dataset may look very different to those who understand how the data were produced.
Centralisation can therefore reduce organisational distance while increasing cognitive distance. The decision is brought closer to the institutional centre, but the relevant knowledge may remain dispersed across the edges of the organisation. The resulting architecture separates the location of authority from the location of understanding.
This does not mean that local knowledge is inherently superior. Proximity creates visibility, but it also creates limitations. Local actors may understand their immediate environment exceptionally well while being unable to see wider patterns, comparative evidence or systemic consequences. A central unit can identify inequities across regions that no individual region can observe, recognise duplicated effort, detect strategic inconsistency or understand interactions among programmes that appear independent locally. Centralisation can therefore expand some forms of institutional knowledge even as it increases distance from others.
The important distinction is between SCALE OF AUTHORITY and SCALE OF KNOWLEDGE. There is no reason to assume that the two naturally coincide. Some questions require national comparison, others depend on local interpretation, and many require both. The cognitive challenge arises when the institution designs the first while assuming that the second will automatically follow.
This is especially visible when organisations centralise in response to inconsistency. Different local units may make different decisions in apparently similar situations, leading the centre to conclude that discretion itself is the problem. Sometimes it is. Unequal treatment, arbitrary practice or weak implementation standards can justify stronger common control. But variation in decisions can also reflect differences in conditions that are invisible in central representations. If authority is centralised before the institution understands why the variation exists, a genuine knowledge signal may be interpreted as administrative noise.
The result can be a peculiar form of institutional blindness. The centre sees a cleaner system because decisions have become more uniform, while the periphery continues to encounter differences that the central decision architecture no longer represents effectively. Local actors may then compensate informally, reinterpret central rules in practice or repeatedly escalate exceptions. What appears from the centre as implementation failure may partly be a symptom of knowledge that has not been incorporated into the place where authority now resides.
More reporting does not automatically solve this problem. A centralised system can demand increasingly detailed information from local units and still fail to recover the knowledge that decentralised judgement previously used. Reporting requires experience to be translated into predefined categories, and those categories inevitably select what can travel. If the centre does not know which contextual distinctions matter, it may request large quantities of data while excluding precisely the information needed to interpret them.
This is why the question “Does the centre have the data?” is weaker than the question “Can the decision architecture access the knowledge relevant to this judgement?” The first concerns informational availability. The second concerns cognitive reach.
The distinction also complicates familiar debates about centralisation and decentralisation. These are often treated as choices about where authority should sit, but institutional cognition suggests that authority and knowledge can have different optimal distributions. An institution may have legitimate reasons to centralise a decision while preserving distributed sensing, interpretation and expertise. Conversely, it may decentralise some decisions while retaining central analytical capabilities that help local actors understand system-wide conditions. The cognitive architecture does not have to reproduce the authority architecture exactly.
Artificial intelligence could make this separation more important rather than less. Centralised AI systems can process information from across an organisation at a scale unavailable to individual decision-makers, potentially allowing central institutions to detect patterns that were previously invisible. This may reduce some cognitive disadvantages of centralisation. Yet a model trained on centrally standardised data can also amplify the distance between formal representation and situated knowledge if important contextual distinctions never enter the system. Computational reach is not the same as contextual understanding.
AI may also make centralisation appear cognitively safer than it is. If a central decision-maker receives a comprehensive dashboard, predictive assessment or generated synthesis, the resulting informational richness can create the impression that relevant knowledge has been successfully concentrated. But completeness of presentation does not demonstrate completeness of cognition. The system may contain everything that was captured centrally while remaining unaware of what the institutional architecture never learned how to represent.
A more cognitively mature approach therefore begins by mapping not only where decisions are made but where the knowledge required by those decisions resides. Some of that knowledge may be held in central datasets, some in specialist units, some in local practice and some in relationships with communities or external organisations. The objective is not necessarily to relocate all of it. Often that would be impossible, and sometimes undesirable. The challenge is to ensure that the decision can reach the knowledge without requiring the knowledge to abandon the context that makes it meaningful.
This changes the design question. Instead of asking only whether authority should be centralised or decentralised, institutions can ask what forms of knowledge each decision requires, where those forms are generated, how they can enter judgement and which contextual properties must survive the journey. A centralised decision can then remain cognitively connected to distributed knowledge rather than pretending that central authority has made distributed knowledge unnecessary.
The same principle protects against romanticising decentralisation. Moving decisions closer to local knowledge may reduce cognitive distance, but it can also weaken access to knowledge that exists elsewhere in the system. Local proximity does not provide systemic visibility automatically. The problem is therefore not solved by choosing one end of an organisational spectrum. It is solved by recognising that the distribution of cognition and the distribution of authority are related but distinct design problems.
Centralising decisions can decentralise knowledge because authority moves more easily than understanding. When decision rights migrate towards the centre while relevant knowledge remains distributed across local environments, professional communities and operational experience, the cognitive distance between knowing and deciding can increase. The challenge for an intelligent institution is not to make authority and knowledge occupy the same organisational location, but to build a decision architecture capable of reaching the knowledge it needs wherever that knowledge resides.