The question systems thinking answers is why competent people working hard reliably produce an outcome none of them chose. Its answer is that the outcome is a property of the arrangement rather than of the people inside it, and that the arrangement will keep producing it until the arrangement changes.

Where the idea comes from

The term covers a body of work assembled between the 1940s and the 1970s. Ludwig von Bertalanffy proposed a general system theory (1968) on the argument that organisms, organisations and machines share formal properties the specialised sciences had no vocabulary to describe. Norbert Wiener's CyberneticsWiener, N. (1948). Cybernetics: Or Control and Communication in the Animal and the Machine. MIT Press. The word is from κυβερνήτης, steersman — the governing of a course by information about departure from it. (1948) supplied that vocabulary: feedback, control, and the regulation of a process by information about its own output. W. Ross Ashby (1956) added the constraint that has proved hardest to evade — the law of requisite variety, that a controller must command at least as much variety as the system it regulates.

Jay Forrester turned the apparatus on organisations. Industrial Dynamics (1961) modelled firms as stocks, flows and delays. His 'Counterintuitive Behavior of Social Systems' (1971)Forrester, J. W. (1971). 'Counterintuitive Behavior of Social Systems.' Technology Review 73(3). The source of the term policy resistance: a system absorbs a well-aimed intervention and returns to its prior state. draws the conclusion that still does most of the work — that complex systems present obvious intervention points which are reliably the wrong ones. Donella Meadows, who had worked on those models, later ranked intervention points by the leverage they carry ('Leverage Points', 1999), placing goals and paradigms far above parameters, where nearly all managerial attention is spent.

Two formulations from the management side are worth holding. Russell Ackoff (1981): the performance of a system depends on how its parts interact, not on how they act taken separately, so improving every part in isolation can degrade the whole. And Stafford Beer (2002): the purpose of a system is what it doesBeer, S. (2002). 'What is cybernetics?' Kybernetes 31(2). Abbreviated POSIWID. Aimed at organisations that describe themselves by their charter: a hiring process that states it selects for competence and in fact selects for availability has availability as its purpose, whatever the charter says. — an instruction to read intent from output rather than from stated objectives, and the sharpest diagnostic in the literature.

The mechanism: feedback and delay

A feedback loop is a structure in which a system's output returns to influence its own input. Reinforcing loops amplify a change: customers produce referrals which produce customers. Balancing loops resist one: hiring to raise output raises coordination cost, which lowers output per head. Most attempts to change an organisation fail because they push against a balancing loop nobody identified, which is Forrester's policy resistance stated locally.

Delay separates cause from effect in time. Cutting marketing raises margin this quarter and empties the pipeline two quarters later, by which point the cut is no longer a candidate explanation. Delay is what allows a bad decision to look good long enough to be repeated, and it is why second-order effects are systematically underweighted: the first-order effect arrives while anyone is still watching.

Local optimisation

The most common systems failure is improving a part and degrading the whole. Eliyahu Goldratt's theory of constraints (1984) gives the cleanest statement: throughput is set by the bottleneck, so effort spent anywhere else raises local utilisation and changes nothing measurable. Every function can hit its target while the system underperforms, and every scorecard will be green. Goodhart's law is the same structure seen from the measurement side. Conway's law (1968) is its architectural case — an organisation ships a design that copies its own communication structure, whatever design it intended.

The objections

Karl Popper (1957) argued that holistic reform is unfalsifiable and practically unbounded, and that only piecemeal intervention can be corrected by its own failures. The charge lands: a systems account can always accommodate a failed intervention by widening the boundary, and an explanation that cannot fail is not doing work.

Herbert Simon (1962) supplies the sharper technical limit. Most complex systems are nearly decomposableSimon, H. A. (1962). 'The Architecture of Complexity.' Proceedings of the American Philosophical Society 106(6). Interactions within subsystems are of a higher order than interactions between them, which is why hierarchy is the common form of complex systems and why analysing a subsystem in isolation usually works.: interactions inside a subsystem are far stronger than interactions across its boundary. Where that holds, examining parts separately is correct and systems thinking is overhead. The live question is when near-decomposability fails, not whether holism is generally superior.

The third objection is empirical. William Nordhaus (1973), reviewing the world model behind The Limits to Growth (Meadows et al., 1972), titled his critique 'World Dynamics: Measurement Without Data' — the parameters had been asserted rather than estimated. A stock-and-flow diagram is a hypothesis, not evidence, and the method's standing risk is that it produces satisfying explanations too cheaply. Peter Checkland (1981) made the adjacent point from inside the tradition: hard systems methods assume an agreed problem definition, which human organisations do not supply, and his soft systems methodology treats that definition as the contested object rather than the input.

What it rules out

It rules out character as the explanation of a recurring outcome. It rules out a green local metric as evidence that the system improved. It rules out the stated purpose of an arrangement as evidence of its actual purpose.

It does not rule out individual accountability. Structure explains recurrence, not every instance, and the distinction is where the idea is most often abused: invoking the system for a single failure converts an explanation into an alibi. The test is repetition under different post-holders.

Using it

Three questions, in order. Where in the structure does this behaviour come from — what would have to be true of the arrangement for this to be the normal result? Which loop holds the present state in place, and is it reinforcing or balancing? And what happens two steps after the intervention, at the constraint, rather than one step after it here? Inversion is the useful companion: ask what would reliably produce the outcome you are getting, and check whether you have built it.

Sources

Ackoff, R. (1981). Creating the Corporate Future. Wiley. · Ashby, W. R. (1956). An Introduction to Cybernetics. Chapman & Hall. · Beer, S. (2002). 'What is cybernetics?' Kybernetes 31(2). · Bertalanffy, L. von (1968). General System Theory. Braziller. · Checkland, P. (1981). Systems Thinking, Systems Practice. Wiley. · Conway, M. (1968). 'How Do Committees Invent?' Datamation 14(4). · Forrester, J. W. (1961). Industrial Dynamics. MIT Press. · Forrester, J. W. (1971). 'Counterintuitive Behavior of Social Systems.' Technology Review 73(3). · Goldratt, E. (1984). The Goal. North River Press. · Meadows, D. (1999). 'Leverage Points: Places to Intervene in a System.' Sustainability Institute. · Meadows, D. et al. (1972). The Limits to Growth. Universe Books. · Nordhaus, W. (1973). 'World Dynamics: Measurement Without Data.' Economic Journal 83(332). · Popper, K. (1957). The Poverty of Historicism. Routledge. · Senge, P. (1990). The Fifth Discipline. Doubleday. · Simon, H. A. (1962). 'The Architecture of Complexity.' Proc. American Philosophical Society 106(6). · Wiener, N. (1948). Cybernetics. MIT Press.