Institutions often speak about data as if its meaning were simply waiting to be extracted. A number appears in a dashboard, a trend emerges in an administrative series, a survey reports a change in public behaviour, and the expectation is that the information itself will indicate what is happening. This expectation is understandable because good data can constrain speculation and reveal patterns that would otherwise remain invisible. Yet even perfectly accurate information does not contain its own explanation. Between observation and understanding lies an interpretive task that institutions cannot eliminate simply by improving measurement.

Consider a statistic showing that applications for a public service have fallen by twenty per cent. The number may be entirely correct, but its significance remains open. Demand may genuinely have declined. Eligibility rules may have changed. Citizens may have found alternative services. A digital process may have become harder to use. Regional offices may be recording cases differently. The population itself may have changed. A temporary event may have disrupted access. The observation tells the institution that fewer applications were recorded; it does not tell the institution why that happened or what the decline should mean for policy.

This distinction is easy to lose because numerical precision can create an impression of semantic precision. A figure expressed to one decimal place looks exact, but exactness in measurement is not exactness in interpretation. The same observation can support different institutional meanings depending on the context in which it occurs. A ten per cent increase in emergency demand may indicate deteriorating social conditions in one place, successful identification of previously unmet need in another, or merely a change in reporting practice somewhere else. The data can remain unchanged while its significance changes with geography, history, timing and institutional circumstance.

Context therefore does not sit outside information as optional background. It participates in the production of meaning. A rise in unemployment has one significance during a recession, another during a period of rapid labour-market restructuring and another when measurement categories have recently changed. A delay of two weeks may be intolerable in an emergency licensing process and irrelevant in a long-term planning programme. A budget underspend may indicate efficient delivery, weak demand, implementation failure or unrealistic initial forecasting. Observations become institutionally intelligible only when they are connected to the conditions that make them significant.

Historical context is particularly important because institutional data often compare present conditions with a past that is treated as if it were a neutral baseline. Yet the baseline itself may have been produced under different legal rules, administrative structures, social expectations or technologies. A current indicator can look abnormal simply because the institution is comparing it with a period governed by a different system. Conversely, a stable indicator can conceal substantial change if the meaning of what is being measured has shifted over time.

Geography introduces another layer. National averages can conceal very different local realities, while a finding that appears significant in one jurisdiction may have little explanatory value in another. Institutional arrangements, demographic composition, labour markets, infrastructure and public behaviour all affect what observations mean. The same level of service uptake may represent high accessibility in one context and severe exclusion in another. A number is therefore never just a number once it enters institutional reasoning; it is a number produced somewhere, under particular conditions, for a particular population.

Timing can alter significance just as strongly. Institutions operate across multiple temporal scales, and an observation that looks alarming over a week may be insignificant over a decade. A sudden fall in performance may be an early warning, a temporary fluctuation or part of a recurring seasonal pattern. Without an appropriate temporal frame, data can be accurate and still misleading in practice because the institution attaches significance to variation that the relevant system does not support.

The question being asked also matters. The same dataset can support different interpretations when used for different purposes. A hospital waiting-time measure may be useful for operational management but insufficient for understanding health outcomes. A school attendance indicator can reveal administrative participation without explaining learning quality. Tax-compliance data may show changes in reported behaviour without establishing whether underlying economic activity has changed. Information acquires institutional meaning partly through the relationship between the observation and the question the institution is trying to answer.

This is why interpretation cannot be reduced to technical processing. Statistical models, visualisation tools and artificial intelligence can help institutions detect patterns, compare variables and process complexity. They can improve the speed and scale at which evidence becomes available. But the decision about what a pattern signifies still depends on assumptions about context, mechanism and institutional purpose. A model may identify that two variables move together; it does not automatically establish whether one causes the other, whether a third factor matters or whether the relationship is relevant to the decision being made.

Nor is interpretation simply a matter of subjective opinion. The fact that data require interpretation does not mean that every interpretation is equally plausible. Interpretations can be tested against additional evidence, compared with known mechanisms, examined for consistency and challenged by alternative explanations. Good institutional interpretation is disciplined rather than arbitrary. The important point is that discipline must exist because meaning is not contained transparently within the data themselves.

This becomes especially visible when institutions disagree while using the same evidence. One department may interpret rising expenditure as evidence of growing need, another as evidence of inefficiency and another as evidence that a programme has expanded successfully. The disagreement does not necessarily arise because one group has the facts and another does not. It may arise because each is placing the same facts within a different explanatory frame. The institution therefore needs to distinguish disagreement about observations from disagreement about what those observations mean.

The distinction also changes how institutions think about data quality. Improving accuracy, completeness and timeliness remains essential, but better data quality cannot compensate for weak interpretation. An institution may know with increasing precision exactly what is happening at the level of observable indicators while remaining uncertain about why it is happening or what should follow. Better measurement reduces one kind of uncertainty without automatically reducing all others.

This is where the idea of “evidence-based” decision-making can become misleading if interpreted too mechanically. Evidence does not move directly from database to decision. It passes through selection, framing, comparison and judgement. Which evidence is considered relevant, which baseline is chosen, what counts as an anomaly and which contextual factors are treated as important all shape institutional understanding. Evidence constrains judgement, but it does not remove the need for judgement.

A mature institution therefore asks not only whether its information is accurate but whether the conditions required to interpret that information are visible. It asks what population produced the observation, what historical baseline is being used, whether measurement practices have changed, which institutional structures shape the result and what competing explanations remain plausible. These questions are not additions to data analysis; they are part of what makes data usable for institutional cognition.

The need for interpretation also explains why simply centralising information rarely produces understanding automatically. A shared platform can make multiple datasets accessible to the same users, but access does not settle their meaning. Two indicators can sit next to one another on a dashboard without any explicit account of how they should be related. Technical integration may therefore improve visibility while leaving the cognitive work of interpretation unresolved.

The deeper problem is that institutions can mistake informational certainty for interpretive certainty. They may know exactly what has been recorded and remain unclear about what the record signifies. This is particularly dangerous when precise numbers create institutional confidence that has not been matched by explanatory confidence. Decisions can then become strongly anchored to observations whose contextual meaning has not been sufficiently examined.

Artificial intelligence increases both the opportunity and the risk. AI systems can identify patterns across enormous volumes of information, classify cases and generate summaries at speeds institutions could not previously achieve. These capabilities expand observation dramatically. Yet the greater the volume of processed information, the more important it becomes to distinguish detected regularity from interpreted significance. A pattern generated by an algorithm still has to be situated within the institutional world in which action will follow.

None of this weakens the value of data. On the contrary, it clarifies what data can and cannot do. Data can show institutions aspects of reality they would otherwise miss, challenge intuition and make disagreement more tractable. But they become knowledge only through processes that connect observation to context and significance. An institution that fails to build those processes may become increasingly data-rich while remaining interpretively fragile.

Data does not explain itself because accuracy and meaning are different properties. An observation can be entirely correct while its institutional significance remains uncertain, and the same observation can mean different things across places, times, histories, populations and governing conditions. Institutional understanding therefore begins not when information has merely been collected, but when the institution can place that information within a context that makes its significance intelligible. The question is never only whether the data are right. It is also what, in this particular situation, the data should be understood to mean.