Chapter 7: The Complexity Argument

Here’s draft chapter 7, sent from Bristol this time as our maternity and paternity France/Italy/Switzerland road trip has ended (Corisca was the highlight! 😍). It’s nice to be home though, and hopefully this means the remaining chapters will be landing in your inbox more freqently now…

As always, I’d love to read your thoughts and feedback, as all my drafts are just that: drafts. I’ve received lots of helpful pointers on earlier chapters 🙏

Part III: Why ‘Transformation’ Can’t Scale

Chapter 7: The Complexity Argument

Now we get to the crux of the matter. I’ve been gesturing at complexity for the last few chapters, using it to help explain why change fails. But I haven’t yet laid out the argument in full. What is complexity, specifically? How does it apply to organisations? And what does it actually mean for whether our change and transformation plans are likely to succeed? Complexity is a fact of life, and one we work very hard to ignore, which goes some way to explaining why so many frameworks struggle when they meet reality. Understanding it properly, as opposed to name-dropping the C word willy-nilly, requires giving up on certain hopes about what we can predict and control. That’s scary, which is possibly why I failed to fully understand it for so long, even though Sami had handed me the pigeon on a plate over twenty years ago.

You Can’t Sweep Complexity Under the Carpet

It’s fair to say the field of complexity doesn’t excel at explaining itself to Joe Bloggs, so I’ll lay it out as clearly as I can, drawing heavily on Sami’s book. Firstly, I need to clarify that none of this is Sami’s invention. It’s the accumulated work of scientists across biology, physics, and mathematics who spent the twentieth century figuring out that the so-called clockwork universe had limits. Prigogine showed how systems hold their shape by constantly burning energy. Kauffman gave us fitness landscapes. The Santa Fe crowd of the nineties briefly made complexity famous. Sami synthesised this work in 2002 to help explain why nations don’t develop as per government/UN/World Bank plans, and I’m repurposing it for organisations. Because, unfortunately, those in international and organisational development alike tend to be oblivious to the fact that nations and organisations don’t play by linear rules. I should say that not everyone in the field thinks this move is legitimate. Some complexity thinkers argue human systems need an entirely separate theory, because people have identities and intentions in ways that sandpiles don’t. Perhaps. But nothing people add to a system makes it easier to predict.

So here goes. Complex systems are composed of many interacting parts, and these interactions produce behaviours that can’t be predicted by understanding the properties of the parts alone. It’s not enough to have many parts, linear systems can have many parts too. The parts in complex systems are connected and influencing each other in ways that create feedback loops and emergent, unpredictable patterns. Picture water in a bathtub. With the tap and plug both closed, the water sits perfectly still. That’s order, every part of the system locked into a predictable pattern. Open the tap fully and the water splashes and swirls with no discernible pattern at all. That’s chaos: no pattern, no stability, everything in flux at once; pandemics and financial crashes. But set the tap flowing at just the right rate with the plug removed, and something else happens. The water finds a rhythm. New water constantly enters, old water constantly drains, and yet the system as a whole settles into a stable, recognisable pattern that nobody designed. This is the third state, self-organised complexity, where chaos and order interact to produce something that is neither. Everything that follows applies to bathtubs and bird flocks too. Organisations are worse, because their parts have opinions.

Organisations can occupy any of the three states, and can contain all three at once. Parts of any organisation should be still bathwater. Payroll should be boring; nobody wants an emergent, self-organising approach to being paid. But suffocating bureaucracy is what happens when the whole organisation is run that way: everything locked into place, nothing adapting. An organisation in genuine crisis is the tap fully open. Bureaucracy can be endured for decades. Crisis burns energy faster than the organisation can replace it, so it doesn’t last. It resolves into some new pattern, or it finishes the place off. But taken as a whole, organisations function best in the third state, because taken as a whole they are made of people who interact constantly, forming relationships and making decisions that affect each other. As much as consultancies and top brass like to pretend otherwise, it’s these interactions that shape the emergence and evolution of patterns, cultures, norms, ways of working. Suggesting they are designed and implemented from above is fantasy. And much like the bathwater, an organisation can look stable on the surface while everything within it is in constant motion, right up until a small blockage or a sudden burst of flow tips the whole thing into a different state. There’s no telling in advance which nudge will do it. The technical term for systems like this is “complex adaptive systems”. Complex because of the emergent behaviours. Adaptive because the parts can learn and change in response to their environment. Complex adaptive systems have certain properties that make them behave differently from linear systems.

Telltale Signs

First, cause and effect are not reliably linked. In linear systems the connection holds. Drop the ball, it falls. In complex systems the link comes loose. A small change, a new hire or a casual comment from a leader, can cascade through the system and produce massive effects. A major intervention, an expensive restructure or transformation programme, can be absorbed with no effect at all. And the same cause can produce different effects each time it occurs. This is why transformation programmes so often disappoint. The intervention seems large and significant, yet the system absorbs it. Endless meetings happen, the new structure is unveiled, and a few months later everything seems much as it did before. The effort wasn’t disproportionate to the outcome; the two were never reliably connected in the first place.

Second, the whole is not the sum of its parts. Patterns, behaviours, and properties arise from the interactions of parts that don’t exist in the parts themselves, often out of the blue. Culture is emergent. It arises from how individuals interact, not from anything located in one of them. Innovation is emergent too, ideas combining and recombining across multiple minds rather than issuing from a single one. So is dysfunction. It arises from interaction patterns, not from individual bad actors. The behaviour emerges from relationships, and relationships can’t be reduced to the properties of the people in them.

Third, taking the system apart doesn’t reveal much about how it behaves as a whole. This is Chapter 4’s reductionist pillar, inverted. A clock can be understood through its gears and springs. An organisation can’t be understood through its org chart, its processes, or its people examined one at a time. You can know everything about the individual parts (and people) in an organisation and still not be able to predict how the organisation will behave, because the behaviour lives in the interactions, not the components. This should trouble anyone in the diagnosis business. Every audit, every diagnostic, is an exercise in decomposition, and decomposition is precisely the method that doesn’t work here. It’s also a snapshot, which is the second problem. To know what a complex system is doing you have to observe it in real time. All of it. This would be laborious work and it wouldn’t invoice well. So the industry sells the snapshot instead, a picture of what the system looked like the week the consultants were in the building. I’ve sold a few of those myself. The slide decks were glorious.

Fourth, outcomes cannot be predicted in advance. This follows from the first three, but it deserves its own statement because it’s the property the change industry most needs to deny. The system is one thing; what we can know of it in advance is another. In linear systems, if you do X, you get Y. In complex systems, if you do X, you might get Y, or Z, or nothing, or something completely unexpected. The outcome depends on the state of the system at the moment of intervention, the history that led to that state, and countless other factors that can’t be fully known or controlled. This is why best practices don’t transfer reliably. A practice that worked brilliantly in one organisation might fail completely in another, even if both organisations look similar on paper.

Something else follows from these. Complex systems are path dependent. Where the system ends up depends on where it’s been. History plays a leading role. Two organisations that look identical today might behave completely differently because they have different histories: different founding stories, different crises survived. The past is embedded in the present in ways that constrain and enable what’s possible. This is why you can’t just copy successful organisations. Even if you could replicate their current practices perfectly, you couldn’t replicate the history that made those practices work. The practices emerged from a specific path; transplanted to a different path, they become something else. Complex systems only make sense in hindsight. You can trace backwards and the story holds together, the founding decisions, the market conditions, the people who stayed, the wins that shaped the culture. But none of it was available in advance. You couldn’t have predicted it, and any attempt would have changed the path.

There’s another way to see this. Imagine organisations existing on terrain such as hills and valleys. A peak means you’re well suited to your current conditions: your market, your people, your history, your constraints. A valley means you’re struggling. This terrain, in complexity parlance, is your fitness landscape. The linear assumption is that there’s one universal landscape with one peak that every organisation should aim for. Find the best practices and climb to the summit. What worked over there will work over here. Organisational development doesn’t work like that. It isn’t a rush to the nearest summit but a leisurely exploration of the possibilities.

What’s more, the landscape isn’t universal. Each organisation exists on its own terrain, shaped by its own coevolving systems: competitors, markets, regulators, suppliers, internal teams all adapting to each other simultaneously. Your landscape is different from mine. Another organisation’s peak might be your valley. Their terrain was shaped by their sector, their history, their particular people, their specific constraints. You can’t transplant yourself onto their landscape. You’re stuck on your own. Worse, the landscape shifts. What’s a peak today can become a valley tomorrow. The conditions that made something successful change. Markets move, people leave. Today’s thriving creature is tomorrow’s dodo.

Survival of the Most Stable

The question the landscape leaves open is how anything survives on it, let alone improves. Sami’s book keeps returning to a cycle. An evolving system, whether an animal staying alive, a company struggling to grow, or a whole nation seeking development, has to keep moving through cycles of survival, adaptation and learning. To adapt successfully, a system first has to survive long enough to evolve into its next stable pattern. Survival of the fittest, he wrote, is really survival of the most stable. Not the strongest, not whichever virtue this year’s operating model promises. The most stable. Stability is what buys the time in which adaptation can happen at all.

And stability doesn’t come free. The second law of thermodynamics says that things left to their own devices drift towards disorder; a pile of rubble never assembles itself into a building. Keeping a system in a stable state takes continual energy from outside, which in an organisation means the often tricky work of people interacting, day after day, holding a recognisable pattern together while the details churn beneath it. Too little of that and the second law wins; the pattern dissolves. Too much and the system is killed the other way, held so rigidly that nothing new can happen. Learning is the third part of the cycle. The system takes in its environment and spots regularities. It responds in ways that help it survive long enough for the next round of adaptation. Impair any part of that and the whole process suffers. This holds for every complex adaptive system, from immune systems to economies. Organisations aren’t exempt because they have a strategy document.

Big-bang transformation works on the opposite assumption. The programme’s logic is that the current stable pattern is the problem, so the pattern must be broken: restructure, redefine the operating model, all at once, to a deadline. Sometimes the pattern really is the problem. But it is also the thing doing the surviving, the thing that holds the organisation together while adaptation happens, and breaking it is the only part of the plan that reliably works. Nobody can specify what comes next. Deliberately destabilising at scale and at speed attacks the part of the cycle everything else depends on. People stop learning and start bracing. Energy that went into small daily adjustments goes into protecting positions and updating CVs. The programme damages the organisation’s ability to change, then blames the people for resisting.

No Interactions, No Progress

Complex systems tend towards self-organisation. This is the third state of the bathwater. Given the right conditions, patterns emerge without anyone designing them. Flocks of birds, schools of fish, markets, ecosystems all exhibit self-organised behaviour. No one is in charge. No one is coordinating. And yet coherent, stable patterns arise. Organisations can self-organise too. This is, in fact, what the self-management movement was betting on. The bet is that if you remove the constraints of traditional hierarchy, people will naturally organise themselves into effective patterns. The bet isn’t wrong, exactly. Self-organisation does happen. But it doesn’t happen reliably or predictably. The conditions that allow it to emerge are context specific and often fragile.

Sami was clear about what those conditions are. For self-organisation to emerge, people must be free to interact with each other, not constrained by rigid hierarchy, silos, or fear. They must be capable of interacting, which means having the skills, the time, and the information to actually engage with each other meaningfully. And there must be simple guiding principles that command broad support. Basic shared understandings about how things work here, not elaborate rulebooks that nobody follows. There’s a distinction here that “simple guiding principles” doesn’t quite capture. In ordered systems, constraints determine what must happen. The rules on an assembly line aren’t suggestions; they’re the mechanism. In complex systems, constraints work differently. They shape what can happen without specifying what does. The philosopher Alicia Juarrero worked this out in the late 1990s, a few years before Sami’s book appeared, and Dave Snowden has taken the work further since, distinguishing governing constraints, which set limits on what happens, from enabling constraints, which allow things to happen without determining them. The conditions are the space within which something might emerge, not a prescription. That’s why you can’t install them from outside.

These conditions aren’t the point in themselves. They exist for the sake of what goes on inside them, the interactions between people, because that is where the energy comes from. Volume is not the measure. Plenty of organisations are wall to wall with interaction, meetings all day and messages all night, and starving all the same, because the interactions carry no trust and nothing much else. Hold most organisations up against the conditions and you’ll see why self-management is rare. Freedom to interact is usually partial, granted within limits somebody else sets. Capability is patchy, since people are overloaded, and often short of the information they’d need to contribute. Shared principles either drown in a policy manual (or handbook if trendy) nobody reads or never get agreed at all. I’ve written a few of those manuals. Any one of these can be present. All three at once, holding steady long enough for something to emerge, is the rare part. Sami put it plainly. No interactions, no progress.

Feuds and Full Calendars

One suppressor deserves naming on its own. Conflict. Sometimes it’s live, a feud everyone has learned to work around. Sometimes it’s buried, a disagreement so carefully avoided that half the team’s conversations are shaped by not having it. Either way the effect is the same. People give each other a wide berth. Conversations that should happen don’t. The interactions thin out, and the organisation runs on less, usually while everyone involved insists things are fine. Leaders tolerate this more often than they’d admit, because the person at the centre of it is a friend, or frightening, or the conversation is one they’d rather not have. And the damage outlasts the conflict. Interactions run on trust accumulated slowly over time, and trust spent doesn’t refill on demand. Even once the air is cleared, people can keep avoiding each other out of habit, and the interactions take time to return. There are no shortcuts back to healthy human interactions. And while they’re gone, the cost is bigger than a bad atmosphere. Starved of the energy that interactions supply, the cycle of survival, adaptation and learning is broken. The team swings between flare-ups and frozen silence, chaos and order, everything except the state in between, self-organised complexity, where adaptation happens.

If interactions supply the energy, it would seem to follow that more connectivity is always better. It isn’t. Connectivity has a working range, like the tap. Too little and the system settles into order, silos sealed, nothing reaching anything else. Too much and everything is coupled to everything, and the system tips towards chaos. When every team’s work depends on every other team’s agreement, conflicting requirements can only be resolved by compromise all round, and the compromises shed variety. Teams that once suited different conditions converge on the one shape everyone can live with, and the organisation ends up fit for exactly one set of circumstances. And circumstances change. Events, dear boy… There’s a name for this. Complexity catastrophe, where progress grinds to a halt under repeated local failures, everyone busy and nothing getting better. The everyday version is not exotic. It’s the calendar. When the meetings required to coordinate the work outnumber the hours available to do it, that’s excessive connectivity, experienced one dreaded invitation at a time.

Negotiating with a Hurricane

These conditions can’t be conjured up the way consultants typically work. You can’t manufacture freedom to interact by running workshops on psychological safety. You can’t create capability with a new meeting format, or establish shared principles by writing them on a poster. That’s treating conditions like software you can install. What you can do is different, and more limited. You can remove what suppresses interaction, the policies that punish honesty, the leaders who hoard control. You can stop doing what drains energy, the pointless reporting and the performative meetings. And you can create space where organic patterns might emerge, though you can’t force them to. This work is slow, indirect, and often fails. The conditions might never develop, or they might fade when circumstances change. You’re removing obstacles and hoping something better grows in the space you’ve cleared. That’s a very different promise from transformation.

Recall the four properties. Cause and effect are not reliably linked. The whole is not the sum of its parts. Taking the system apart tells you little about how it behaves. And outcomes cannot be predicted in advance. Systems like this can’t be controlled the way linear ones can. You can nudge them, you can try to create conditions in which better outcomes might emerge, but there’s no formula for it, and you can’t engineer desired results with confidence. The system is too sensitive to conditions you can’t fully know or control. Path dependence and the tendency to self-organise deepen the trouble for anyone trying to steer. It’s already shaped by where it’s been, and it will organise itself around whatever you do, in patterns of its own making rather than yours. This is the core of the argument. Organisations are complex adaptive systems, and therefore they don’t respond to intervention the way linear systems do. The linear toolkit of diagnose, plan, implement, and measure assumes a level of predictability and control that complex systems don’t offer. It’s the wrong tool for the job, like trying to negotiate with a hurricane.

Not Systems Thinking, Not Chaos

A word on what this isn’t. Systems thinking sounds like the same territory, and the two get used interchangeably. They are not the same. Systems thinking keeps the designer in the picture. Map the feedback loops, locate the levers, the points of maximum influence, intervene where the map says. Its best thinkers knew the limits, and said so. But the practice sold under the label kept the promise intact. Understand the system well enough and you can steer it. Complexity thinking is more sceptical, and for damn good reason. The map is always partial, the levers won’t stay put, and the act of intervening changes the system being intervened in. Read this book as systems thinking and you’ll leave with a subtler way to steer, which is precisely the move it’s warning you against. The design-and-control assumption survives by redecorating. The opposite misreading is just as available. Complexity doesn’t mean chaos. Complex systems aren’t random; they have regularities, patterns that recur and can be observed. Weather is the textbook nonlinear system, famously hard to predict, yet short-term forecasts are still useful. It’s the ten-day forecast you shouldn’t plan a wedding around. Organisations have regularities too, and we can work with them, run safe to fail experiments, and adapt as we learn. Just don’t fall for the idea that any system this alive can be understood well enough to be steered.

Chapter 4 set out what I believed, that the right practices would reliably produce better organisations, that the patterns found in progressive organisations could be extracted and transplanted, that practice was the bottleneck. Those beliefs assume a reliable relationship between intervention and outcome, and that understanding the parts (the practices) gives you control over the whole (the organisation). Complexity theory says otherwise. The patterns in progressive organisations might be effects rather than causes. Emergent properties of specific conditions, not transportable solutions. This doesn’t make the practices worthless. They’re ingredients, not recipes. Whether the possibilities they open up are realised depends on history, culture, personalities, timing, luck, none of which can be controlled or fully predicted.

Understanding this earlier would have changed how I talked about the work and tempered my expectations about what could be achieved. It would have prepared me, and the people I worked with, for the messiness and unpredictability that are inherent in complex systems. But taking complexity seriously means giving up the hope that the right approach will produce reliable results. Accepting that failure isn’t always fixable, that sometimes the conditions just aren’t right. Sitting with uncertainty rather than fighting it with confident frameworks. Most people in the transformation business aren’t willing to do this. The business model depends on confidence, on having answers. Complexity undermines all of that. So it gets acknowledged superficially, yes, organisations are complex, yes, change is hard, and then ignored in practice. The frameworks keep getting sold and the results keep disappointing. I was part of this pattern for years. I knew Sami’s work and I could explain, albeit clumsily, the difference between linear and complex systems. But when it came time to sell my services or course, I reverted to linear assumptions. The market demanded it. Clients wanted solutions, not philosophy lectures about the limits of intervention.

Partial Answers

The question I couldn’t answer then, and still struggle with now, is what it would actually look like to take complexity seriously in practice, as a genuine orientation towards the work rather than a framework. What would you do differently? What would you promise? What would you refuse to promise? I don’t have complete answers. But I have some partial ones, developed through years of getting it wrong.

You’d be direct about uncertainty, and not in the “we can’t guarantee outcomes” legal-disclaimer sense. I mean admitting you don’t know what will happen. Clients don’t enjoy hearing it, but they’re still better served by it than by false confidence. You’d also ask what kind of problem you’re facing before deciding how to act. Some of any organisation is still bathwater, and still-water problems can be planned and implemented with certainty. The payroll fix doesn’t need a hypothesis. The moving water is different. Everything there runs on interactions, nobody can call the outcome in advance, and anything you try is an experiment whether you admit it or not. So you’d admit it. You’d measure what actually happens, not whether the thing got delivered, and adapt as you learn. Your role would be a modest one. You’re not the one who changes the organisation. The people inside it do that, or don’t. At best, you can create some space, introduce some ideas, maybe shift some conditions slightly. But the work is always theirs, not yours. Most of what you try won’t produce dramatic results, and you’d have to accept that. The patient, modest work of creating slightly better conditions might do more good than the big transformation programme. But of course, it’s harder to sell and harder to take credit for. That won’t change until leaders genuinely understand complexity. This book is my small nudge at those conditions.

Experiments are the learning leg of the cycle. Learning is how a complex adaptive system finds its way to its next stable pattern, and an experiment is learning made deliberate. It’s also the only form of intervention that doesn’t pretend to know how things will turn out. An experiment can fail, and often will. So sizing is the discipline. Because a system has to survive long enough to adapt, experiments must be sized so that failure never threatens survival. Each one small enough that failure costs little, and enough of them that some will land. The moment an experiment becomes too big to fail, it has stopped being an experiment and become a transformation programme with better branding.

Anyone can run an experiment, and it will usually teach the person who ran it something. But without the freedom and capability to interact, and without shared principles to make sense of what happened, the system can’t read the result. The learning stays where it landed, one person or one team slightly wiser, and travels no further. Interactions are how learning moves, and where they’re suppressed, even a good experiment is a letter nobody delivers.

The change industry has heard all this, and it has a product. The pilot portfolio. Success criteria agreed up front, RAG statuses reported monthly, a steering group deciding which pilots scale. It’s a plan wearing a lab coat. The outcomes are specified before anything runs, failure is a red status to be explained away rather than information to be used, and whatever learning occurs travels upwards to the steering group instead of outwards through the organisation. And any process that promises to reliably deliver the right experiments is the recipe moved one level up.

These orientations don’t add up to a methodology. They can’t be packaged and certified. They’re more like a stance, a way of approaching the work that takes complexity seriously instead of paying it lip service. Whether this stance is commercially viable, I don’t know. The market for tentative uncertainty is smaller than the market for confident solutions. But complexity is insistent. You can ignore it, but the tap keeps running and the water keeps finding its own pattern, whatever the slide deck says.

---