Large institutions have good reasons to standardise. Common definitions make information easier to compare, shared formats allow systems to exchange data, common procedures reduce uncertainty about how work should be performed, and standard categories make it possible to coordinate activity across organisational boundaries. Without some degree of standardisation, complex institutions would spend enormous amounts of time translating among their own parts. Every transfer of information could require reinterpretation, every joint decision could begin with negotiation over terminology, and every attempt to combine knowledge could encounter incompatible structures. Standardisation reduces this friction, and institutional capability often depends upon it.

Yet the very differences that standardisation removes are not always meaningless variation. Some encode local knowledge, professional distinctions, alternative ways of representing a problem or sensitivity to conditions that a common framework does not capture. When these differences are eliminated indiscriminately, an institution can become easier to coordinate while becoming less capable of perceiving the complexity of the world it governs. The central challenge is therefore not whether to standardise, but what should become common and what should remain different.

The appeal of standardisation is easy to understand because its benefits are highly visible. If every department records the same phenomenon differently, institutional comparison becomes expensive. If one unit measures age in categories, another as a continuous variable and another through eligibility groups, combining their records may require substantial transformation. A common data standard can solve the problem. If several agencies use different definitions of the same administrative status, coordination can become unreliable. Agreeing on a shared definition may dramatically improve interoperability. In such cases, variation produces little cognitive benefit and substantial operational cost.

Problems emerge when the institution assumes that all variation has the same character. A difference can be noise, historical accident or unnecessary duplication, but it can also be information. Two professional communities may distinguish cases differently because different properties of those cases matter for the decisions each community must make. Two local authorities may collect different contextual information because the populations they serve present different conditions. A specialist unit may preserve a category that appears redundant from the centre because the distinction predicts an operational consequence that disappears when the data are aggregated.

This is why VARIATION ≠ NOISE. The existence of difference tells us nothing by itself about whether that difference should be eliminated. Institutions need to understand what a distinction does before deciding whether harmonisation represents improvement.

Consider a national organisation trying to standardise how local offices classify vulnerable households. A common framework could make comparisons easier, support national resource allocation and reduce arbitrary variation in service provision. These are substantial advantages. Yet local offices may also have developed distinctions that capture forms of vulnerability specific to their environments: transport isolation in rural areas, housing instability in expensive urban areas, seasonal employment in particular regions or linguistic barriers within particular communities. If national standardisation removes those distinctions because they are not universally applicable, the resulting system may become more comparable while becoming less informative where those local conditions matter.

The trade-off is not between rational central standards and irrational local variation. It is between different cognitive functions. A national classification makes certain patterns visible across the whole system; local distinctions make other patterns visible within particular environments. An intelligent architecture asks how much commonality is necessary for knowledge to travel without assuming that everything must be represented identically before it can be connected.

Professional standardisation presents the same difficulty. Multidisciplinary institutions often want specialists to use common language so that their knowledge can circulate more easily. This can be valuable, particularly when differences in terminology serve no substantive purpose. But specialist vocabularies sometimes preserve conceptual distinctions that general language cannot express efficiently. Replacing them with a common institutional vocabulary may improve apparent communication while reducing the precision of the knowledge being communicated.

The objective is therefore not maximum semantic uniformity but sufficient interoperability. Different communities need enough common structure to recognise and use one another’s knowledge, while retaining the distinctions necessary for expertise. COMMON FORMAT ≠ COMMON UNDERSTANDING, just as different formats do not necessarily imply incompatible understanding.

This becomes especially important when standardisation shapes what the institution can record. A standard form or database schema does more than organise information after it has been produced. It influences which observations can enter institutional memory at all. If every case must fit the same fields, information that does not correspond to those fields may become difficult to preserve. Over time, the institution can lose not only local variation in administrative practice but evidence that alternative representations of the problem ever existed.

Standardisation can therefore alter institutional cognition indirectly. Once a common classification becomes embedded across systems, reporting and evaluation, the institution increasingly encounters reality through that classification. Differences that were eliminated for coordination purposes may disappear from datasets, analytical models and eventually institutional attention. A technical decision about interoperability can thus become a cognitive decision about which distinctions the organisation remains able to perceive.

This does not make standardisation inherently dangerous. The opposite problem is equally serious. An institution that preserves every local distinction indefinitely can become incapable of comparing experience, transferring learning or coordinating action. Knowledge may remain rich but trapped. Each unit develops a representation perfectly adapted to its own environment, while the institution lacks a common structure through which those representations can interact. Cognitive diversity without interoperability can become fragmentation.

The real design problem therefore lies between two failure modes. Too little standardisation can make institutional knowledge mutually unintelligible. Too much can make institutional knowledge artificially uniform. One produces friction and isolation; the other produces ease at the cost of cognitive variety. Neither extreme provides a sufficient architecture for institutional intelligence.

This is why LESS FRICTION ≠ MORE INTELLIGENCE. Reducing coordination costs is valuable, but cognitive quality depends on what is lost in the reduction. If two categories can be merged without affecting any meaningful judgement, standardisation has removed unnecessary complexity. If merging them eliminates a distinction that matters for understanding causes, consequences or appropriate action, the same operation has removed knowledge.

Artificial intelligence makes this trade-off particularly visible. AI systems often benefit from standardised data structures, labels and taxonomies. Common formats can improve training, evaluation and deployment across organisational environments. Yet forcing heterogeneous institutional realities into a single schema can conceal differences that matter. A model trained on harmonised categories may perform efficiently while reproducing the assumptions embedded in the harmonisation process. The institution gains computational interoperability but may lose representational diversity.

AI can also work in the opposite direction. Instead of requiring every source to become identical before it can be combined, computational systems may increasingly help translate among heterogeneous representations. If so, institutions may be able to achieve some of the benefits of interoperability without eliminating as much variation beforehand. But technological translation does not remove the design question. Someone still has to decide which differences are substantive and which are merely incompatible formats.

That judgement cannot be made purely technically because the value of a distinction depends on institutional purpose. A category that is irrelevant for one decision may be essential for another. A local variation that complicates national reporting may preserve evidence necessary for local intervention. A professional distinction that appears excessively detailed to a generalist may carry consequences that only become visible in specialist practice.

For this reason, good standardisation should be selective and reversible where possible. Institutions can establish common cores while allowing extensions, maintain shared definitions while preserving contextual qualifiers, or create translation layers between systems rather than forcing every system into one representation. The specific mechanisms will differ, but the cognitive principle remains constant: interoperability should remove unnecessary incompatibility without automatically removing meaningful difference.

This also changes how institutions should evaluate standardisation projects. Success cannot be measured only by adoption rates, reduced processing time or the percentage of systems using a common schema. Those measures capture important operational benefits, but they do not reveal what forms of knowledge may have become harder to express. Evaluation should also ask which distinctions disappeared, whether those distinctions carried useful information, and whether alternative perspectives remain recoverable when circumstances require them.

The broader lesson is that cognitive diversity is not the opposite of institutional coherence. An institution can maintain several legitimate ways of seeing a problem while still possessing enough shared architecture to act collectively. Indeed, the ability to preserve useful difference while making knowledge interoperable may be one of the characteristics of a mature institutional cognitive system.

Standardisation is therefore best understood as an instrument rather than an objective. Its value depends on what kind of friction it removes and what kind of difference it preserves. Used carefully, it can allow knowledge to travel, comparisons to become possible and coordination to become dramatically easier. Used indiscriminately, it can make the institution internally smoother while narrowing the range of distinctions through which it understands reality.

Standardisation can make coordination easier and thinking poorer because not every difference is an obstacle. Some variation is redundant, but some carries knowledge. The task of an intelligent institution is to distinguish between the two. It should standardise where difference creates unnecessary incompatibility while preserving the distinctions that allow specialised, local and alternative forms of understanding to remain available. The goal is not maximum uniformity or maximum diversity, but an architecture in which knowledge can travel without requiring the institution to forget useful ways of seeing the world.