‍ ‍Goldratt’s Challenge, Forty Years On: The Missing Mechanism

Nearly forty years ago, Eliyahu Goldratt posed a challenge that remains remarkably contemporary. His argument began with a simple observation about hierarchical management. All managers are decision-makers. What distinguishes them is the subject and magnitude of the decisions they are expected to make.

The implication is significant. An organisation may possess a single strategic intent, but it does not execute as a single decision-maker. Execution occurs through a hierarchy of distributed decisions made across functions, processes, value chains and organisational levels.

Goldratt recognised the inherent difficulty. Decisions that are entirely rational within one part of an organisation do not necessarily combine to produce rational behaviour for the organisation as a whole. The optimum of the parts is not the optimum of the total -Goldratt.

His challenge was to find the appropriate rules. Rules that enable people to make sensible local decisions while causing those decisions to remain coherent with the objective of the whole. He made an important connection. In organisations, those rules are often expressed through what we call measurement.

The question nearly forty years later is whether transformative change has really risen to that challenge.

The mechanism in the guts of the computer

One of Goldratt’s most striking observations concerned an assumption forming around information systems. Somewhere in the “guts of the computer”, there appeared to be an expectation that a mechanism would exist capable of translating broad managerial direction and increasingly detailed organisational data into appropriate operational instructions. The statement now seems unusually prophetic.

The expectation was not necessarily that computers would develop intelligence in the contemporary sense. It was subtler. That increasingly sophisticated information and automation would somehow supply a deeper guiding mechanism between enterprise purpose and local action. Data would become direction. Information would become instruction. Broad managerial intent would somehow become appropriate operational behaviour.

The subsequent history of enterprise technology can almost be read as a succession of attempts to fulfil that expectation. ERP provided integration. Advanced planning provided optimisation. Business intelligence provided visibility. Automation provided speed and consistency. Artificial intelligence now provides increasingly sophisticated interpretation, recommendation and autonomous action.

The capabilities inside the computer have changed almost beyond recognition. Yet Goldratt’s underlying question has not.

What is the appropriate rule?

The computer can execute a rule. It can enforce one, optimise against one, detect deviations from one and increasingly infer an appropriate action within one. But none of these capabilities establishes why the governing rule itself should contribute to the strategic purpose of the organisation.

 The missing mechanism was never simply computational. It was the management logic connecting organisational intent to appropriate distributed action. That logic has to be designed.

Measurement is part of that mechanism

This requires a broader interpretation of measurement. Measurement is conventionally treated as observation. It tells management what happened, whether a target was achieved and whether performance improved.

But consequential measurements do something more. They influence what receives attention. They establish what constitutes acceptable performance. They shape priorities, define tolerances and legitimise intervention. They influence the choices that appear rational to the person making the decision. Measurement does not merely describe the operating system. It participates in governing it.

System redesign and measurement redesign are treated as connected but distinct activities. The operating design establishes the conditions expected to produce an outcome, while measurement provides the structure through which those conditions and the resulting system behaviour can be observed, interpreted and corrected.

Goldratt’s challenge can be stated more precisely. What rule should govern this decision, and why should rational behaviour against that rule contribute to the objective of the whole? That is a causal question. We have measurement systems. What we often lack is causal derivation.

Most organisations do not suffer from a shortage of measures. They have financial measures, operational KPIs, management scorecards, benchmarks, utilisation measures, service measures, dashboards and increasingly extensive analytical environments.

The problem is that these measurement systems are usually assembled rather than derived. Measures accumulate from accounting convention, functional practice, regulatory requirements, industry benchmarks, existing systems and inherited management structures.

Individually, many are perfectly reasonable. But a collection of reasonable measures does not necessarily constitute a coherent control system. If every manager behaves rationally according to the measurement governing their local decisions, what behaviour will the organisation as a whole produce? That question exposes Goldratt’s problem in contemporary form.

A measure may be locally relevant and globally damaging. A local target may improve while enterprise performance deteriorates. And the individual manager need not have misunderstood the strategy or acted irrationally. They may simply have optimised against the rule the organisation provided.

The problem is not one of alignment. It is one of causal coherence.

Strategic intent must undergo causal treatment

If the appropriate rules cannot simply be selected from a KPI catalogue, they have to be derived from somewhere. The logical starting point is strategic intent. Strategic intent describes a desired future state. But embedded within it is also a causal proposition.

If particular conditions can be created and maintained, then particular system behaviours should emerge, and those behaviours should contribute to the intended strategic outcome. Once the proposition is declared the reasoning can progress.

Strategic intent -> causal hypothesis -> required system behaviour ->  necessary operating conditions -> required controls  ->governing decisions->  decision rules   ->measurement

This sequencing fundamentally changes the role of measurement. The measure is no longer selected because it appears relevant to the strategy. It is derived because a particular condition must be observed.

That condition matters because a decision may be necessary. The decision matters because it influences a required system behaviour. And the behaviour matters because it forms part of the causal hypothesis through which strategic intent is expected to become an outcome.

Measurement becomes the operational expression of causal reasoning.

KPIs must be derived rather than selected, because generic measures describe what organisations commonly observe rather than what a particular organisation needs to observe to execute its chosen strategic position. Hierarchy then creates a harmonisation problem. Causal derivation is necessary, but it is not sufficient.

The rules have to operate through an organisational hierarchy. Different managerial levels necessarily make different decisions. Procurement, operations, logistics, service, planning and finance do not require identical measurements.

Uniform measurement is neither possible nor desirable. What is required is harmonisation. Vertically, a local rule should remain causally traceable to strategic intent. Horizontally, acting rationally against one rule should not systematically undermine the conditions being maintained elsewhere in the system.

This changes the test for an appropriate measurement. If the decision-maker behaves rationally according to this measure, what consequence should that behaviour have for the wider system, and what is the causal basis for believing that consequence supports strategic intent?

This is where Goldratt’s local-optimisation argument becomes more than an operations principle. It becomes a test of organisational design. The problem is not local decision-making. The problem is local decision-making governed by rules whose system-level consequences have not been causally established.

Measurement must also follow authority

There is another implication that deserves greater attention. A measurement only becomes an effective control mechanism when it reaches a decision-maker capable of changing the condition it reveals. Information without authority is reporting. Information connected to the appropriate decision authority is regulation.

This provides an important design principle for hierarchical measurement. Each organisational level does not need visibility of every measure. It needs visibility of the conditions it has the authority and capability to regulate. The hierarchy s attenuate complexity rather than replicate it.

Strategic leaders require information about strategic outcomes and material changes in assumptions. Design authorities require evidence concerning whether the conditions supporting the design remain valid. Operational managers require signals concerning the conditions they can directly influence.

The measurement architecture becomes a distributed control system rather than an expanding dashboard. Measurement ownership must sit with those capable of changing the condition revealed by the measure, and scaling measurement through the organisation should be governed by decision authority rather than simple replication.

The cybernetic tradition helps clarify what this architecture must accomplish. Ashby established that effective regulation requires the ability to distinguish relevant system states and possess sufficient response variety to deal with the disturbances that matter.

Beer subsequently carried that logic into organisational design, showing that sensing, communication and regulation need to recur throughout a viable organisation.

The relevance to Goldratt is direct. If decision-making is distributed, sensing and control must also be distributed. What should be sensed and regulated? It has to come from the causal reasoning connecting strategic intent to required system behaviour.

Only then can the organisation determine what condition matters, how frequently it should be observed, what constitutes material deviation and what response should follow. That is the bridge between causal reasoning and organisational control.

The mechanism must work in both directions

The measurement system cannot merely derive rules downward from strategic intent and then assume those rules remain correct. Organisations operate in open, changing environments. The causal assumptions underlying the original rules must remain provisional. This means the architecture must operate in two directions.

Strategic intent -> causal reasoning -> operating conditions -> rules -> distributed

Observed behaviour -> evidence -> interpretation -> assumptions -> design->  strategic reconsideration.

This is where feedback becomes essential. Operational feedback must return to the reasoning and assumptions embedded in the design so that tactics, control settings and, where necessary, strategic assumptions can be reconsidered.

Without that second direction, the organisation may become highly disciplined at executing a rule that is no longer appropriate. Feedback is what makes the organisation adaptive

The timing of feedback also matters.

A measure that identifies deterioration after the outcome has already occurred may be useful for explanation or accountability. It is not useful for control. For measurement to regulate behaviour, observation, interpretation and response must occur while a meaningful correction remains possible.

Control depends upon the relationship between measurement cadence, decision latency, authority and the correction window. Adaptation requires the capability to sense a material change, interpret its significance against the causal design, determine whether intervention or redesign is required, act while a correction remains possible and observe the system response.

The measurement architecture must support both control and learning. It regulates according to the organisation’s current causal hypothesis. And it supplies the evidence required to challenge that hypothesis.

Rising to Goldratt’s challenge

This brings us back to the mechanism Goldratt expected management to find. The missing mechanism is a causal control architecture. Its rules are derived from strategic intent. Those rules are harmonised through the organisational hierarchy so that local rationality supports system rationality.

Measurement is placed at the level at which corresponding decision authority exists. Feedback operates recursively so execution can challenge the assumptions from which the rules were originally derived. And cadence and decision latency ensure that information arrives while there is still time to act.

Begin with strategic intent as the governing reference condition. Treat  intent as a causal hypothesis. Required system behaviour is reasoned from it. Necessary and sufficient conditions are identified. Controls and decision rules are derived. The measurement architecture is then designed to make those rules visible, actionable and testable throughout the organisational hierarchy.

The result is not simply traceability from strategy to KPI. It is a persistent causal argument explaining why each important measurement exists, which condition it observes, which decision it informs and how that decision is expected to contribute to the intended outcome. Feedback then runs the argument in reverse.

If the expected behaviour does not emerge, the organisation can interrogate the decision, the operating condition, the control rule, the causal assumption and ultimately the strategic intent.

Measurement becomes part of the mechanism through which the transformation works, rather than something added afterward to determine whether it worked.

Forty years later

The remarkable aspect of Goldratt’s challenge is how little it has aged. The guts of the computer have become extraordinarily capable. The hierarchy still needs locally actionable rules that remain coherent with the objective of the whole.

Local optimisation remains a predictable consequence when those rules are disconnected. And the feedback needed to reconsider those rules as circumstances change remains considerably less developed than the technology used to execute them.

AI makes the contrast particularly visible. For the first time, the technology inside the computer can increasingly observe, interpret, recommend and act. But greater computational intelligence does not eliminate the need for governing logic.

It increases its importance. Goldratt challenged us to find the appropriate rules.

Nearly forty years later, the response may be becoming clearer. Causally derive the measurement and control architecture from strategic intent. Harmonise it through the organisational hierarchy. Connect measurement to decision authority and use recursive feedback to continually test whether the rules and the assumptions supporting them remain valid.

The intelligence in the computer’s guts is arriving. The unfinished work is designing the management system it should execute.