Organisations as Complex Adaptive SystemsWhy the Machine Model Keeps Winning

In 1911 Frederick Winslow Taylor published The Principles of Scientific Management, and its best-remembered passage concerns a man loading pig iron at Bethlehem Steel. Taylor calls him Schmidt. By timing each movement, prescribing each rest and paying for compliance, Taylor reports raising the man’s daily load from around twelve and a half tons to forty-seven. The book states its principle early and without embarrassment: in the past the man has been first; in the future the system must be first.

More than a century later, organisations still speak Taylor’s language. They have operating models and levers, pipelines and cascades. They count people as headcount and full-time equivalents. They draw themselves as boxes joined by lines and, when something goes wrong, go looking for the broken part. The machine is the most durable metaphor in management, and it describes almost nothing about how organisations behave.

The better description has been available for decades: organisations are complex adaptive systems. The interesting question is why that description, which most people recognise the moment they hear it, has made so little difference to how organisations are run.

Part of the answer lies in what metaphors do. In Images of Organization, first published in 1986, Gareth Morgan argued that every theory of organisation rests on a metaphor, and that every metaphor is a way of seeing that is also a way of not seeing. Call an organisation a machine and you will notice its inputs, outputs and inefficiencies with great clarity. You will struggle to notice anything it does that you did not design it to do. The Fractured Lens framework begins from the same suspicion of any single metaphor, which is why it looks through four forces at once. Long before anyone draws a process map, the metaphor has already decided what management is able to perceive.

What a Complex Adaptive System Is

The clearest working definition comes from Paul Plsek and Trisha Greenhalgh, writing in the BMJ in 2001 about why health care kept defeating the people trying to manage it. A complex adaptive system, they wrote, is a collection of individual agents with freedom to act in ways that are not always totally predictable, whose actions are interconnected so that one agent’s actions change the context for other agents. Their examples were the immune system, a termite colony, the financial market, and just about any collection of humans: a family, a committee, a primary care team.

Plsek and Greenhalgh were blunt about where the machine metaphor breaks: it fails badly wherever the parts of a situation are neither constant, nor independent, nor predictable, which describes almost everything involving people. Each clause of their definition quietly dismantles the machine. Agents with freedom to act are people who interpret and improvise. Actions that are not always predictable mean the same input will not reliably produce the same output. Interconnection means that every intervention alters the ground it lands on. A new metric changes what people attend to, which changes what they say to each other, which changes what the metric ends up measuring.

From those properties come the behaviours that researchers in complex systems science have spent decades describing. Small causes can have large effects while large causes vanish without trace. Order arises from local interactions that nobody designed. Boundaries are fuzzy, because people belong to several systems at once. And the system has a history, so what worked last year can fail this year for reasons nobody can point to.

Human systems add one property that termites lack. People notice how the system describes them and respond to the description. Label a team underperforming and the label becomes part of the team’s situation: some members work harder to escape it, and the best of them update their CVs. Announce that a function is strategic and watch the queue of people wanting to join it. The act of describing an organisation changes the organisation being described, which means managers are never simply observing the system they run. They are among its most influential agents, and their maps are part of the territory.

Most people who have worked in an organisation have watched this happen without the vocabulary for it. The clearest example is the project status report. A project is reported green for months, then amber for a fortnight, then suddenly red, and everyone privately knew it had been in trouble since spring. The reporting system assumed a machine: accurate signals travelling upward from parts to operators. What it met was a system of people who read the room, protected each other, understood what red would cost them and adapted accordingly. The report did exactly what a complex adaptive system does with a control instrument. It learned to satisfy it.

The same adaptation happens inside individual working lives. Professionals learn to translate judgement into process so that it can be seen and counted: the clinician documents for the audit, and the engineer writes the ticket in the form the dashboard can read. Much of this is sensible. Some of it slowly replaces the work it was meant to describe, until people spend more of their week proving they are functioning parts than doing the thing the part was for. Nobody decided this. It emerged, from thousands of reasonable responses to a model that can only value what it can measure.

Why the Machine Model Keeps Winning

If the machine is so poorly matched to organisational life, why does it persist? The usual answer is inertia: managers were trained on it and have not caught up. That answer is too generous. The machine model survives because it provides two things organisations value more than accuracy.

The first is relief from the discomfort of not knowing. A machine can be understood. Its parts can be listed and its failures traced. A board approving a budget and a manager explaining a variance both need to speak as though intervention A will produce outcome B. The machine lets them. Strategy documents and transformation roadmaps are written in its grammar, and when the world declines to follow the plan, the plan is revised while the grammar survives. The annual planning cycle shows the ritual most clearly. Each year the organisation produces a plan that assumes it can predict the next twelve months, and spends much of those months explaining why it could not. The variance is analysed, the forecast restated, the plan rolled forward. As The Strategy Delusion explores, the exercise persists because it provides a shared performance of foresight, and that performance matters to the organisation whether or not the forecast comes true. Organisations are built to manufacture certainty, and the machine metaphor is where it gets made.

The second is a settled arrangement of power. Someone operates a machine. There is a designer and there are parts, a control room and a floor. The metaphor ratifies the hierarchy that uses it: if the organisation is a machine, the people at the top are its engineers, and their authority rests on the claim that they understand how it works. Describing the organisation as a complex adaptive system concedes that nobody fully understands it, including the people paid most to understand it. That concession is costly, and the people who would bear the cost are the same people who decide which metaphor the organisation uses. The power to describe the organisation and the power to run it sit in the same hands.

Together these explain why failure never seems to discredit the model. When a change programme stalls, the post-mortem finds poor implementation or resistance. The failure is located in the parts, and the parts are retrained or replaced. As The Alibi of Certainty argues, the search for someone accountable protects an organisation from the harder question of whether its understanding was ever adequate. Each failure becomes further evidence that the machine needs better parts.

Emergence in Organisations: The Organisation Nobody Designed

Inside every organisation runs another one that nobody drew. It lives in the informal networks through which information actually travels, in the workarounds that keep broken processes alive, in the unwritten rules about who can be contradicted and who cannot. Complexity researchers call this emergence: pattern arising from interaction rather than instruction.

Emergent order is often more reliable than the designed kind. New starters learn it within weeks, and they learn it from each other rather than from the induction pack: which policies are enforced and which are decorative, and whose approval matters whatever their title. The formal organisation predicts what should happen. The emergent one predicts what will. Its fragility shows when a long-serving administrator retires: processes that appear in no documentation quietly stop working, and the organisation discovers how much of it had been running through one person’s memory and relationships.

This is also where meaning is made. People do not experience their work through the org chart. They experience it through relationships and stories, through the constant and mostly silent interpretation of what is going on and what it means for them. Karl Weick called this sensemaking, and it is one of the main ways a complex adaptive system adapts: each person reads the situation and acts on the reading, and the action becomes part of the situation everyone else reads. Ralph Stacey pushed the point further, describing organisations as patterns of conversation, reproduced and transformed in ordinary exchanges between the people in them. On that view, a strategy is only as real as the conversations it manages to change.

And this is where belonging is decided. Formal membership comes with a contract. Actual membership comes from being inside the networks where things are known before they are announced. Two people with the same title can hold entirely different places in the emergent organisation: one consulted before a decision, the other informed after it. The machine model cannot register the difference, because in a machine two identical parts are identical. People can register it very precisely, and much of what gets called disengagement is simply the experience of being addressed as a part while living as a person.

The language gives it away. A person becomes a resource and a team becomes a cost centre. None of these words is malicious, and each is useful in its place. Their accumulated effect is a steady message about what kind of membership is on offer: you belong here to the extent that you are interchangeable. People hear that message even when nobody intends to send it. Some respond by withholding the parts of themselves that do not fit a job description. Others hold on to the emergent organisation more tightly, because the informal networks are the only place their knowledge, history and judgement are still recognised. When a restructure breaks those networks in the name of efficiency, it removes the very thing that was keeping the machine running.

Interventions designed for the machine behave strangely once they reach the living system. A restructure redraws the formal lines and leaves the emergent organisation to reorganise itself around them, often reproducing the old patterns under new names. A values programme announces what the organisation stands for, and the emergent organisation quietly tests each value against what gets rewarded and keeps the ones that survive. The system absorbs the intervention and adapts. That is its nature.

When Complexity Becomes an Alibi for Power

It would be satisfying to end with complexity as the enlightened answer and the machine as the error it corrects. The story has a sting in it.

A few months after Plsek and Greenhalgh’s article, the BMJ printed a reply from a doctor, Ian Reid, under the title Let them eat complexity: the emperor’s new toolkit. Reid argued that the language of complexity could let managers treat uncomfortable situations, including tensions caused by poorly managed services, as mysterious natural phenomena to be contemplated rather than addressed. He was objecting to one example in the original article, and his warning has aged better than most of the enthusiasm it answered.

Complexity can serve as cover. When a leader says the organisation is a complex system and outcomes could not have been predicted, that may be an honest admission. It may also be a way of disowning a decision whose consequences were predictable, and were in fact predicted, by the people nearest to the work. Emergence can describe genuine self-organisation. It can equally describe abandonment, with people left to sort out among themselves what nobody above them would decide, and the result credited to their autonomy.

Complex adaptive systems are also far from level. Some agents change the context for everyone else far more than others. An email from the chief executive rearranges more of the system than a concern raised on the front line, and a budget decision constrains thousands of local interactions before any of them happen. The vocabulary of emergence can smooth this asymmetry away, presenting an outcome as the product of the whole system when a few parts of the system shaped most of it. When an organisation stops describing itself as a machine, power becomes harder to locate, and power that is hard to locate is also hard to question.

Complexity also changes who is allowed not to know. A chief executive who tells the board that the system is complex and outcomes are emergent sounds thoughtful. A team leader who gives the same answer to a request for a delivery date sounds evasive, and a graduate who says it sounds unprepared. The permission to acknowledge uncertainty travels upward, while the obligation to produce certainty travels down. Those lower in the organisation still owe estimates, forecasts and commitments to people who may be publicly embracing unpredictability. Whether uncertainty is heard as wisdom or as weakness depends heavily on the standing of the person expressing it, which says as much about who belongs in the conversation as about what anyone knows.

Complexity has even been repackaged for the machine. Frameworks such as Cynefin were built to help people sense what kind of situation they are in, and are regularly used as sorting devices instead: identify the domain, apply the matching method, move on. The complex domain becomes one more box on the diagram. The insight that some situations cannot be controlled gets converted into a procedure for controlling them, and the machine absorbs its own critique, which may be the most complex-adaptive thing it does.

None of this makes the machine model redundant. Nobody can run an organisation without some working fiction of predictability, and the anxieties that fiction soothes are real. Payroll still has to run, and somebody still has to decide what happens on Monday morning. The question worth holding is what the fiction costs, and who is allowed to say, before the report turns red, that the system has stopped behaving as drawn.

Somewhere in every organisation there is a pattern holding things together that has never appeared on a diagram. The people who keep it going usually know exactly what it is. Whether anyone with the authority to redraw the diagram has ever asked them is another matter.