When we say that an institution has learned, we often imagine learning as though it were a single kind of change: something happens, the organisation absorbs the lesson and its future behaviour improves. In practice, however, institutional learning can penetrate an organisation to very different depths. Experience may lead to a small adjustment in procedure, but it may also change what the institution measures, what it pays attention to, whose knowledge matters, where authority is located or how reality itself is represented. These changes are not equivalent, even when all of them can reasonably be described as learning. The depth of institutional learning therefore depends less on how visible the resulting change appears than on which part of the cognitive architecture has been modified by experience, because an institution that changes a procedure has learned something different from one that changes the machinery through which future problems will be perceived, interpreted and judged.

The most immediate form of learning often occurs at the level of procedure. An organisation encounters a problem, identifies a practical weakness and changes what people are expected to do next time. A form is redesigned, an approval stage is added, a checklist is modified, an escalation rule is clarified or a workflow is adjusted. Such changes can be valuable and should not be dismissed as superficial simply because they are procedural. Institutions need reliable ways of converting experience into better routines, and many failures recur precisely because obvious procedural lessons are never embedded. Yet procedural learning normally operates within an existing interpretation of the problem: the institution changes the response while largely preserving the categories, assumptions and relationships through which the situation is understood.

Learning can penetrate further when experience changes what the institution measures. Measurement is not merely a technical activity placed downstream from institutional understanding; what an organisation chooses to count helps define which aspects of reality become visible enough to govern. If repeated experience reveals that existing indicators systematically miss an important dimension of performance, changing those indicators can alter future institutional cognition because different evidence will now enter decision processes. A public service that previously measured only processing speed, for example, may discover that rapid completion conceals repeated errors or unequal outcomes and begin measuring quality, recurrence or distribution as well. The procedural consequences may initially be modest, but the cognitive significance is greater because experience has changed part of the apparatus through which the institution observes itself and its environment.

A related but distinct form of learning occurs when the institution changes what it pays attention to. Organisations receive far more information than they can meaningfully process, so cognition depends not only on what information exists but on what becomes salient enough to influence judgement. An institution may already possess data about complaints, delays, local variation or unusual cases without treating those signals as worthy of sustained attention. Experience can alter that allocation of attention, turning what was previously peripheral into something institutionally significant. This kind of learning reaches deeper than the addition of information alone because the institution has changed the practical hierarchy through which some signals become cognitively consequential while others remain background noise. What matters is no longer simply that new evidence is available, but that the institution has learned to notice something it previously tended to overlook.

Learning can penetrate still further when experience changes relationships between the people and units through which institutional knowledge is produced. Many organisational failures cannot be traced to an absence of expertise inside individual departments; they emerge because relevant forms of knowledge remain separated, arrive too late or fail to modify one another. An institution may therefore learn that its previous division of cognitive labour was inadequate even if every formal unit was performing competently. The resulting change might involve creating new interfaces between frontline staff and policy designers, linking operational and analytical teams more closely, or establishing mechanisms through which external knowledge can enter internal judgement. What has changed in these cases is not merely a workflow but the relational architecture through which institutional cognition is assembled. Experience has taught the organisation something about who needs to think with whom.

Authority represents another possible depth of learning because institutions can discover not only that information was missing or poorly connected, but that the ability to act was located in the wrong place. A frontline unit may repeatedly recognise a problem without possessing authority to respond, while a central body possesses authority without sufficient contextual knowledge to exercise it intelligently. If experience leads the institution to redistribute decision rights, escalation powers or discretion because it has recognised this mismatch, learning has entered the architecture of authority. Such a change can be especially consequential because it modifies the relationship between knowing and acting: the institution is no longer simply processing better information within the same distribution of power, but reconsidering where judgement should become consequential.

Deeper still, experience may alter how the institution represents reality. Every organisation operates through categories, classifications, models and conceptual distinctions that simplify a world too complex to encounter directly. These representations determine what counts as a case, a risk, a success, an exception, a vulnerable population or a policy problem. When experience reveals that one of these representations is inadequate, institutional learning can involve changing the conceptual map itself. A programme may discover, for instance, that a population previously treated as a single category actually contains groups whose circumstances require fundamentally different interpretations, or that what had been classified as individual non-compliance is better understood as a recurring pattern generated by the design of the service. The resulting learning reaches beyond procedure because the institution has altered the conceptual objects upon which its procedures operate.

These forms of learning should not be understood as a simple ladder on which every institution must progress from one level to the next. A procedural change can be exactly the right response to a procedural problem, just as altering authority would be unnecessary if the existing distribution of decision rights is already appropriate. Depth is therefore not a synonym for quality, and deeper learning is not automatically better learning. The relevant question is whether experience has reached the part of the institution that actually generated or sustained the problem. If a recurring failure is caused by an outdated procedure, changing that procedure may constitute sufficient learning; if the procedure repeatedly fails because the institution is measuring the wrong thing, changing the procedure alone will leave the deeper source untouched. The purpose of distinguishing depths is not to establish a hierarchy of prestige but to improve diagnosis.

The same event can also generate learning at several depths simultaneously. A crisis may reveal that an escalation procedure is inadequate, that warning indicators are poorly designed, that senior leadership pays attention to the wrong signals and that local staff possess knowledge that cannot influence central decisions. Responding intelligently may therefore require procedural, measurement, attentional, relational and authority changes at the same time. These changes belong to a common learning episode, but they modify different components of institutional cognition. Treating them all simply as “lessons learned” conceals precisely the architecture that matters most: where experience entered the institution, what it changed and how those changes will affect future judgement.

This perspective also helps explain why institutional learning can remain incomplete even when lessons have been formally recorded. A post-event review may correctly identify what went wrong, yet the insight can remain cognitively shallow if it never modifies the mechanisms through which future situations will be encountered. An organisation can document that early warning signals were missed while continuing to measure the same indicators, allocate attention in the same way and preserve the same relationships between those who see emerging problems and those who hold authority. In such cases the institution possesses an articulated lesson without having allowed that lesson to penetrate very far into its operating cognition. The depth of learning becomes visible only when experience leaves a durable trace in the structures that generate future perception and judgement.

Artificial intelligence makes this question increasingly important because AI can intervene at several of these depths at once. It may simply automate an existing procedure, but it can also introduce new measurements, redirect institutional attention, alter relationships between professional groups, influence where practical authority sits or embed representations of reality into models and classification systems. The important question is therefore not merely whether an institution “learns to use AI”, but which elements of its cognitive architecture are being changed as AI becomes part of institutional work. A technically sophisticated transformation may remain relatively shallow if it accelerates existing procedures without altering the assumptions behind them, while a more modest intervention may have profound consequences if it changes what the institution can recognise or which evidence becomes available to judgement.

At the deepest end of this spectrum lies a more fundamental possibility: experience can modify not merely individual components of institutional cognition but the architecture through which those components interact. An institution may learn that its recurring failures are not located in one procedure, indicator, relationship or distribution of authority, but in the way these elements combine to produce judgement. Measurement determines what becomes visible; attention determines which visible signals matter; relationships determine which forms of knowledge can encounter one another; authority determines which judgements can become action; and representation determines the conceptual world within which all of these processes operate. When learning reaches this level, the institution begins to revise not just particular decisions or routines but the machinery from which future decisions will continue to emerge.

Understanding learning in this way changes the question institutions should ask after experience. Rather than stopping at whether a lesson was identified or whether behaviour subsequently changed, they can ask where the lesson entered the cognitive architecture and how far its consequences travelled. Did experience change only a procedure, or did it reveal that the institution was measuring the wrong thing? Did it add information, or did it change what people considered worthy of attention? Did it improve coordination, or did it expose a deeper mismatch between knowledge and authority? Did it correct a particular judgement, or did it reveal that the categories through which the institution understood the problem were themselves inadequate? These questions make learning more demanding to analyse, but they also make it more concrete because depth becomes something that can be located rather than merely asserted.

Institutional learning can therefore reach from the adjustment of a specific routine to the modification of the architecture that generates institutional cognition itself. No single depth is automatically desirable, and not every experience should trigger fundamental transformation. What matters is whether the change reaches the mechanism that needs to learn. The deepest institutional learning occurs when experience does more than change what the institution does: it changes some part of the cognitive machinery through which the institution will decide what to notice, how to interpret it, whose knowledge to connect, where judgement should become action and how reality itself should be represented when the next problem arrives.