From Data Analyst to System Builder: Why Every Problem Felt Like a New Battle
I’ve spent the past decade analyzing bioinformatics data. Sequencing runs, single-cell experiments, multi-omics projects, clinical databases. Somewhere along the way, without quite noticing when it happened, my job changed. I stopped being a person who analyzes data and became a person who builds the things that analyze data. I went from data analyst to system builder.
That shift taught me something I now see everywhere, and it started with a complaint I kept hearing.
Junior scientists would come to me exhausted. “Every new dataset is a new battle,” they’d say. A new experiment meant new scripts, new bugs, new late nights, new panic before a deadline. They weren’t lazy or unskilled. Many of them were brilliant. But every project felt like starting from zero.
My answer was always the same: “That’s because you don’t have a system, and you don’t think in systems.”
They’d push back, and honestly, their pushback was fair. “Every dataset is different! Every experiment is different! You can’t build a system for every customized study.”
I agree with them. You can’t write one script that fits every analysis. You can’t develop one protocol that fits every custom experiment. And you can’t predict the future from the past.
But here’s what I’ve come to believe after ten years of building pipelines and training people to use them: you can absolutely build a system, and think in systems, for almost everything you do. Your research. Your career. Your money. Your business. Your life. And even your relationships. The datasets change. The structure underneath them doesn’t have to.
Before any of that makes sense, though, we need to answer a deceptively simple question. What actually is a system? And why does having one change everything?
The book that gave me the clearest language for this is Thinking in Systems: A Primer by Donella H. Meadows. Meadows was a scientist trained in the MIT System Dynamics group and the lead author of The Limits to Growth (1972), one of the most influential (and most argued about) books on global sustainability ever written. She drafted Thinking in Systems in 1993, passed away unexpectedly in 2001, and the manuscript was edited by Diana Wright and published in 2008.
I’ll be upfront with you: this book is not a beach read. It’s full of stock-and-flow diagrams, behavior-over-time graphs, and model equations. It can feel academic and dry. So in this article I’m going to do what I wish someone had done for me: translate its ideas into real stories from biology (my world) and business (my obsession), and show you how to use them.
Let’s start with a toy.
The Slinky Lesson: The Behavior Lives Inside the Structure
Hold a Slinky by its top coils, rest the bottom on your other palm, then pull your lower hand away. The Slinky drops, bounces, and yo-yos up and down.
What made it bounce? Most people say, “Your hand. You moved your hand.”
Now do the exact same thing with the box the Slinky came in. Pull your hand away. The box just hangs there. Nothing happens.
So the bounce wasn’t really caused by the hand. The hand only released a behavior that was already sitting inside the coiled structure of the spring.
That’s the single most important idea in this entire article: a system largely causes its own behavior. Outside events trigger it, but the structure determines what happens next.
When I first absorbed that, I thought about those junior scientists. Their pain wasn’t really caused by each new dataset. The dataset was the hand. The chaos was latent in how they worked: no reusable components, no checkpoints, no shared conventions, no feedback. Every new dataset just released the same bounce.
Hold onto that thought. We’ll come back to it.
What Is a System? The Three Core Components Hiding in Plain Sight
A system is not just any pile of stuff. A system is a set of things interconnected in a way that produces its own pattern of behavior over time, and that achieves something.
That definition hides three ingredients. Once you see them, you can’t unsee them:
- Elements: the parts
- Interconnections: the relationships and flows that link the parts
- Function or purpose: what the whole thing is actually doing
Sand scattered on a road is not a system. Add a handful or take one away, and you still just have sand on a road. But take a player off a soccer team, or remove an organ from your digestive tract, and you no longer have the same system. That’s the difference.
Here’s a quick test I now run in my head whenever I’m trying to figure out whether I’m looking at a system or just a bunch of stuff:
- Can I identify the parts?
- Do the parts affect each other?
- Do the parts together produce something different from what each produces alone?
- Does that behavior persist across different circumstances?
If the answers are yes, you’re looking at a system, and it will behave like one whether or not you understand it.
Elements: The Parts We Notice First (and Overrate)
Elements are the easiest part of any system to see. In a cell, they’re DNA, ribosomes, mitochondria, membrane proteins. In a company, they’re employees, products, offices, and cash. In an analysis pipeline, they’re the scripts, the tools, the reference genomes, and the compute cluster.
Because elements are visible, we obsess over them. Hire better people. Buy better software. Get a faster server.
But here’s the surprising part: changing the elements usually has the least effect on the system.
Think about any research lab. Students graduate. Postdocs move on. Within five or six years, nearly every person can be different. Yet the lab often keeps its personality: the same pace, the same habits, the same strengths and blind spots. Your body is the same way. Your gut lining renews itself every few days and your red blood cells turn over in a few months, yet you go on being you.
A system keeps being itself, even through near-total replacement of its parts, as long as its interconnections and purpose stay intact.
Interconnections: The Invisible Rules of the Game
Interconnections are the relationships that hold elements together. Some are physical flows: nutrients moving through a bloodstream, money moving through a business, files moving from one step of a pipeline to the next. But many of the most important interconnections are flows of information.
In a cell, a hormone binding a receptor is information. In a company, a sales report reaching a manager is information. In a pipeline, a log file telling you a step failed is information.
Change the interconnections and the system can transform completely, even with the same parts. Keep the same eleven soccer players but switch to basketball rules, and you have a whole new ball game. Keep a plant’s cells but reverse its gas exchange so it takes in oxygen and releases carbon dioxide, and it isn’t behaving like a plant anymore.
Purpose: The Least Visible, Most Powerful Part
Purpose is the hardest part to see, and usually the most decisive. And here’s a lesson that stuck with me: you learn a system’s real purpose by watching what it does, not by reading what it says.
A company can print “innovation” on every wall. But if it promotes the people who never take risks and quietly sidelines the ones whose experiments fail, its actual purpose is risk avoidance. A university can claim its purpose is teaching, but if tenure depends almost entirely on publications and grants, the system will reward professors who neglect students to publish.
Purposes also nest inside each other, and they can collide. A student’s purpose might be good grades. A professor’s might be tenure. An administrator’s might be a balanced budget. Each is reasonable. But they can add up to an outcome nobody wanted: students who cheat, professors who ignore teaching, administrators who cut the very faculty the university depends on.
When I think about my own work, this is the part that matters most. The purpose of an analysis pipeline isn’t “run scripts.” It’s “produce trustworthy, reproducible answers to biological questions.” When people forget that, they optimize the wrong thing. They chase speed, or clever code, or a pretty plot, and lose sight of whether the answer is actually right.
So here’s the ranking I now carry around: elements are the most visible and usually the least important. Interconnections matter a lot. Purpose matters most.
Stocks and Flows: The Bathtub That Explains Everything
If you remember only one diagram from systems thinking, make it a bathtub.
A stock is anything that accumulates: water in a tub, money in an account, customers in a business, glucose in your blood, skills in your head, trust in a relationship. A flow is what fills or drains it: the faucet and the drain.
The rules are almost childishly simple:
- If inflows exceed outflows, the stock rises.
- If outflows exceed inflows, the stock falls.
- If they’re equal, the stock holds steady, even though water is constantly moving through it. That’s called dynamic equilibrium.
Simple, right? Yet we get this wrong constantly, because our minds focus on inflows and forget outflows.
Here’s a business example with numbers that are hard to ignore. Most companies pour enormous energy into the faucet: marketing, ads, sales campaigns, customer acquisition. Meanwhile, the drain is wide open. Customers quietly leave and nobody tracks why.
But research by Frederick Reichheld at Bain & Company found that increasing customer retention by just 5% can increase profits by 25% to 95%. Same bathtub. You can fill it by turning up the faucet, or by plugging the drain. Plugging the drain is often far cheaper.
The same logic applies everywhere. A company can grow its workforce by hiring more, or by losing fewer people. You can grow your savings by earning more, or by leaking less. A breakthrough in energy efficiency does to an oil supply exactly what a new oil discovery does.
Your body figured this out long before economists did. Your blood glucose is a stock. Food fills it. Your cells drain it. Insulin and glucagon act like a hand constantly adjusting faucet and drain to keep the level in a narrow, healthy range.
Why Stocks Change Slowly (and Why That’s a Gift)
Here’s the other thing the bathtub teaches: you can turn the faucet instantly, but the water level takes time to change. Stocks change slowly, even when flows change suddenly.
That’s why a forest can’t regrow overnight, why a population takes decades to stabilize, and why a company’s reputation, once damaged, takes years to rebuild. It’s also why your skills don’t evaporate after one bad month, and why one great month doesn’t make you an expert.
Stocks act as buffers, shock absorbers, and sources of momentum. They give you room to experiment and make mistakes without immediate collapse. But they also mean you shouldn’t expect things to happen faster than they can happen. Many people quit right before the stock they’ve been filling starts to show results.
It’s worth remembering every time someone abandons a new habit after two weeks. They turned on the faucet. They just didn’t wait for the tub.
How Systems Behave: Feedback Loops, Delays, and the Dynamics of Change
Once you understand stocks and flows, the next question is: who’s controlling the faucets?
In most systems, the answer is the stock itself. When the level of a stock influences its own inflows or outflows, you get a feedback loop. And feedback loops come in just two basic flavors.
Balancing Feedback Loops: The System’s Thermostat
A balancing loop pushes a stock toward a goal. If the stock goes too high, the loop pulls it down. If it goes too low, the loop pushes it up.
Your home thermostat is the classic example. Room too cold? Furnace on. Warm enough? Furnace off.
As a biologist, I find the body’s version far more elegant. After a meal, your blood glucose rises. Beta cells in your pancreas sense the gap between where glucose is and where it should be, and release insulin. Cells absorb glucose and the level falls. Skip a meal and the opposite loop kicks in: glucagon tells your liver to release stored glucose. Two opposing balancing loops, constantly negotiating, all day, every day, without you thinking about it once.
Balancing loops are sources of stability. But notice: they’re also sources of resistance to change. Every time you try to change something and it stubbornly snaps back (a diet, a company culture, a team’s habits), you’ve likely run into a balancing loop you didn’t see.
There’s a subtle trap here, too. A thermostat set to 68°F in a leaky house will settle slightly below 68°F, because heat keeps draining out even while the furnace is trying to replace it. The same thing happens when you try to pay off a credit card. If your payments don’t account for the interest piling up while you pay, you’ll never reach zero. Your goal has to account for the drain, or you’ll always fall short.
Reinforcing Feedback Loops: Snowballs, Flywheels, and Vicious Circles
A reinforcing loop does the opposite. It amplifies whatever direction it’s already going. The more you have, the more you get. Or the less you have, the less you get.
Compound interest is the textbook case. Money earns interest, the interest earns interest, and the growth accelerates. Here’s a shortcut worth memorizing, the “rule of 70”: divide 70 by the growth rate, and you get roughly how long it takes to double. At 7% a year, your money doubles in about 10 years. At 10%, about 7 years.
Microbiologists watch reinforcing loops every day. Put a few bacteria in rich broth, and each cell divides, and each daughter divides, and the population explodes exponentially. That’s the “log phase” of a bacterial growth curve, one of the first graphs every biology student learns.
In business, the most famous reinforcing loop might be Amazon’s flywheel. Jeff Bezos and his team sketched it out, by most accounts on a napkin, after conversations with management writer Jim Collins around 2001. Lower prices bring more customers. More customers attract more third-party sellers. More sellers mean more selection and a better customer experience. Greater scale drives down costs, which funds even lower prices. Round and round, each turn making the next turn easier.
But reinforcing loops aren’t always virtuous. They also drive vicious circles: soil erosion that kills the roots that hold the soil, price wars, panic selling, arms races, and the spiral of an unhappy team where turnover causes overwork, which causes more turnover.
Once you start looking for these loops, you’ll find yourself asking a new question. Instead of “Did A cause B?” you’ll start wondering, “Could B also be causing A?” In science, that one question prevents an astonishing number of bad conclusions.
Delays: Why Fast Reactions Often Make Things Worse
Now add one more ingredient, and things get interesting: delays.
Information takes time to arrive. Decisions take time to make. Actions take time to have effect. And a balancing loop with a delay in it is almost guaranteed to oscillate.
You’ve felt this in a hotel shower. You turn the handle toward hot. Nothing happens, because of the long pipe. You turn it more. Suddenly it’s scalding. You yank it back toward cold. Nothing. More. Now it’s freezing. You’re not bad at showers. You’re fighting a delay.
Businesses do this at massive scale. In the 1990s, Procter & Gamble’s logistics executives studied the orders for one of their best-selling products, Pampers diapers. Babies, it turns out, use diapers at a pretty steady rate. Yet retailers’ orders swung up and down. Distributors’ orders to P&G swung even more. And P&G’s own orders to suppliers like 3M swung the most. Nobody was irrational. Each player was reacting, with a delay, to the orders of the player below them, adding a little safety margin each time. P&G named it the bullwhip effect: a small flick at the handle becomes a huge crack at the tip. The fix wasn’t working harder. It was sharing real sales data up the chain (P&G moved to vendor-managed inventory with Walmart and 3M), which shortened the information delay.
A more recent example still makes me wince. During the pandemic, demand for Peloton’s at-home bikes exploded. Delivery times stretched out and customer complaints piled up. Management saw a persistent gap between demand and supply and invested heavily in capacity, including plans for a $400 million factory in Ohio. But by the time that capacity was coming online, gyms were reopening and demand was fading. In February 2022 the factory plan was cancelled. By spring 2022, Peloton was sitting on about $1.4 billion of inventory, and its market value had fallen from a peak of nearly $50 billion to under $5 billion.
Here’s the counterintuitive lesson I took from this: when there’s a delay in the system, reacting faster often makes things worse, not better. Overreacting to each new signal amplifies the swings. Sometimes the wise move is to react more slowly and more steadily, or better yet, to shorten the delay by getting better information.
Shifting Dominance: Why Growth Never Lasts Forever
Real systems rarely have just one loop. They have many, pulling against each other, and which loop dominates can change over time. That’s where the most interesting behavior comes from.
Go back to those bacteria. Exponential growth doesn’t last. As nutrients run out and waste builds up, a balancing loop gets stronger and stronger until it overpowers the reinforcing loop. Growth slows, stops, and the culture enters stationary phase. If nothing changes, the death phase follows.
That growth curve in a flask is one of the most important patterns in the world. Systems thinkers call it limits to growth. Every growing thing, whether it’s a population, a startup, a product, an epidemic, or a career, is driven by at least one reinforcing loop and eventually constrained by at least one balancing loop. No physical system can grow forever in a finite environment.
The question is never whether growth will hit a limit. It’s which limit, when, and whether you’ll see it coming.
Why We Need Systems (and Why They Work So Well): Resilience, Self-Organization, and Hierarchy
This brings me back to my junior scientists. Their unspoken question was really: why bother building a system at all? Why not just solve each problem as it comes?
The answer is that well-designed systems do things individual heroic effort can’t. When systems work well, they show three remarkable properties: resilience, self-organization, and hierarchy.
Resilience: The Toyota Fire That Should Have Shut Everything Down
At 4:18 a.m. on Saturday, February 1, 1997, a fire broke out at a factory owned by Aisin Seiki, one of Toyota’s closest suppliers. That factory was the sole source of P-valves, a small but critical brake part used in virtually every Toyota vehicle.
Because Toyota ran a lean, just-in-time system, there were only two or three days’ worth of valves on hand. Analysts predicted Toyota would be shut down for weeks.
Instead, something extraordinary happened. Roughly 200 firms from inside and outside the Toyota group mobilized. Around 70 of them took direct responsibility for producing P-valves, many with no prior experience making them. Engineers worked overtime shifts to improvise production lines. Toyota’s assembly plants reopened after only about two days of complete shutdown.
Researchers who studied the recovery noted that it happened with limited direct control from Toyota, and without fights over proprietary technology or compensation. The network healed itself.
That’s resilience: the ability of a system to bounce back after being stressed. And notice where it came from. Not from one feedback loop, but from many overlapping ones: shared knowledge, long-term trust, regular employee transfers between firms, and a common way of solving problems.
Your immune system works the same way: innate defenses, antibodies, T cells, memory cells, all layered and redundant, so that if one fails, another kicks in.
But here’s the warning hidden in this story. The just-in-time system that made Toyota so efficient is the same thing that left it with only two days of valves. We often sacrifice resilience for efficiency or stability, because resilience is invisible until the moment you need it. A system can look perfectly smooth right up until the day it does something it has done a hundred times before, and crashes.
The right way to think about resilience is as a plateau a system can wander around on safely. A resilient system has a wide plateau with soft, springy edges. As resilience erodes, the plateau shrinks, until the system is balancing on a knife edge without knowing it.
Self-Organization: Simple Rules, Astonishing Complexity
The Toyota recovery also shows the second property: self-organization, the ability of a system to create new structure on its own.
Biology is the master of this. A single fertilized egg, following a handful of genetic and chemical rules, organizes itself into a heart, a brain, and a set of fingerprints no one has ever had before. Bacteria do a smaller version through quorum sensing. Each cell releases a signaling molecule. When the population gets dense enough, the signal crosses a threshold and the whole colony switches behavior at once: forming biofilms, or in the case of Vibrio fischeri, glowing in the light organ of a Hawaiian bobtail squid. No bacterium is in charge. A simple rule produces coordinated behavior.
I see the same principle in the workflow tools I use to build pipelines. In a framework like Snakemake or Nextflow, you don’t hand-write the order of every step. You write simple rules: this output requires these inputs. The engine figures out the whole execution graph by itself, runs independent steps in parallel, and resumes from where it left off if something fails. Simple rules, emergent structure.
Here’s what I wish more organizations understood: self-organization requires freedom, experimentation, and a certain amount of disorder. That makes it scary for managers who prefer control. So systems often suppress it in the name of efficiency, turning creative people into cogs. They get short-term predictability and lose long-term adaptability.
Hierarchy: The Watchmaker Who Never Started Over
The third property is hierarchy, and it’s best explained by a parable that the Nobel-winning economist Herbert Simon liked to tell.
Two watchmakers, Hora and Tempus, each build watches with about a thousand parts. Both are constantly interrupted by customers calling with orders. Tempus assembles each watch in one go, so every time he puts it down to answer the phone, it falls apart and he has to start over. Hora builds stable subassemblies of about ten parts, then combines ten subassemblies into a larger unit, and so on. When he’s interrupted, he loses only a small piece of work. Hora prospers. Tempus goes broke.
That parable describes, almost perfectly, the difference between junior scientists who write one giant thousand-line script per project and the people who build small, tested, reusable modules. The first group is Tempus. Every interruption (a new dataset, a reviewer request, a changed parameter) sends them back to zero. The second group is Hora.
Biology is hierarchical for the same reason: molecules into organelles, organelles into cells, cells into tissues, tissues into organs, organs into you. Each level handles its own business, which keeps any single level from being overwhelmed with information.
But hierarchy can fail in two directions. One is suboptimization: when a subsystem pursues its own goal at the expense of the whole. In biology, we have a name for a cell that stops serving the body and starts multiplying for itself. We call it cancer. In companies, it’s the department that hits its own targets while sinking the company. The other failure is too much central control, the micromanager who second-guesses every decision until the people closest to the problem stop thinking.
The point I keep coming back to is that the original purpose of a hierarchy is to help its lower levels do their jobs better. When the top forgets that, the system begins to rot.
So, why do we need systems? Because a good system bounces back from shocks, adapts without being told, and lets complex work proceed without starting over. That’s exactly what my exhausted junior scientists were missing.
Why Systems Surprise Us, and How to Handle the Surprises
If systems are so wonderful, why do they constantly blindside us?
Because everything we think we know about the world is a model. Our mental models are often good enough. But they leave things out, and the things they leave out are exactly where surprises come from. Here are the blind spots that catch me, and almost everyone I know.
We Watch Events Instead of Patterns and Structure
Turn on the evening news and you’ll hear why the stock market went up or down today. The dollar fell. Interest rates rose. A CEO said something. That’s event-level thinking, and it’s addictive because events are dramatic. But it has almost no power to predict anything or change anything.
One level deeper is behavior over time: the pattern. Not “the market dropped today,” but “volatility has been rising for six months.” Deeper still is structure: the stocks, flows, and feedback loops that produce the pattern.
In data analysis, this should be the first move: before you touch anything, look at the history. Plot the data over time. What’s the shape? Growing? Oscillating? Drifting? The shape of the curve is a clue to the structure underneath. A system that oscillates probably has a delay. A system that grows and then crashes probably has a reinforcing loop running into a limit.
We Think in Straight Lines in a World of Curves
Our brains love linear relationships. Twice the effort, twice the result. Twice the fertilizer, twice the crop. But most of the world is nonlinear.
A little fertilizer helps. More helps less. Too much damages the soil and yields fall. Highway traffic flows smoothly as cars are added, right up until a certain density, then speed collapses to a crawl. Anyone who has run a dose-response experiment knows the sigmoid curve: nothing, nothing, nothing, then a sudden steep jump, then a plateau. And sometimes, past a point, the response reverses entirely.
Nonlinearities matter because they change which feedback loop dominates. They’re why a system can behave one way for years and then suddenly flip into a completely different mode.
We Draw Boundaries That Nature Doesn’t Recognize: The Sparrow Disaster
Every model needs boundaries. We can’t include everything. But we forget that the boundaries are ours, not nature’s. And the biggest surprises hide right at the edges.
In 1958, as part of the Great Leap Forward, China launched the Four Pests Campaign to eradicate rats, flies, mosquitoes, and sparrows. Sparrows were targeted because they ate grain. Citizens banged pots and pans so the birds couldn’t land, until they dropped from exhaustion. A well-known biologist, Zhu Xi, warned that an earlier attempt to wipe out sparrows in Prussia had backfired. He wasn’t heeded.
What the campaign’s boundary left out: sparrows also eat insects, including locusts. With the birds gone, pest populations exploded. A recent study by economists Eyal Frank, Shaoda Wang, and colleagues estimates that the sparrow eradication alone accounted for nearly 20% of the decline in crop production during the Great Famine and was associated with nearly two million excess deaths. In 1960, sparrows were quietly removed from the list and replaced with bedbugs.
This is why I no longer believe in “side effects.” There are just effects. Some of them we wanted, and some we drew our boundary too narrowly to see.
The opposite mistake exists too. Some analysts build models so big, with so many arrows connecting everything to everything, that they bury the actual question. The right boundary depends on the question you’re asking. Draw it for the problem, not for your academic discipline or your org chart.
We Ignore the Limiting Factor (and Forget That It Moves)
Growing grain needs sunlight, water, nitrogen, phosphorus, potassium, and dozens of micronutrients. The 19th-century chemist Justus von Liebig noticed that it doesn’t matter how much nitrogen you add if phosphorus is what’s short. Growth is controlled by whatever is most limiting. Anyone who has grown cells in culture knows the same rule: skip one essential component of the medium and nothing else you add will save the culture.
The tricky part is that the limiting factor changes as the system grows. A young company hires great salespeople, and orders outrun the factory. So they expand the factory, hire fast, and quality drops because training is now the bottleneck. They fix training, orders flood in, and the billing system collapses. Each solution shifts the limit somewhere else.
Genomics went through exactly this. For years the bottleneck was the cost of sequencing. Then sequencing became dramatically cheaper, and the limit moved to computing, storage, and above all, to people who could analyze the data. That shift is the backdrop of the analyst-to-builder journey I described at the start.
Insight comes from looking past what’s abundant today to what will be scarce tomorrow.
We Underestimate Delays
Jay Forrester, the founder of system dynamics, had a rule of thumb for estimating delays: ask everyone involved how long they think it will take, make your best guess, then multiply by three. Every person who has ever planned a software release, a grant, a renovation, or a book is nodding right now.
When there are long delays in a feedback loop, waiting until a problem is obvious means you’ve waited too long. Foresight isn’t optional.
Bounded Rationality: Reasonable People, Terrible Outcomes
Here’s the blind spot that changed how I judge people.
Herbert Simon called it bounded rationality: people make reasonable decisions based on the information they have, but they never have complete information, especially about distant parts of the system. A fisherman doesn’t know how many fish are left or how many his neighbors will catch today. A manager doesn’t know what competitors are planning. So everyone does something sensible from where they stand, and the sensible decisions add up to a result nobody wanted.
The most powerful consequence: if you put a new person into the same position, with the same information and incentives, they’ll usually behave the same way. Firing the person rarely fixes the system.
The good news is that small changes to information can change behavior dramatically. A company called Opower ran large randomized experiments, mailing home energy reports that compared each household’s electricity use with similar neighbors. Across roughly 600,000 households, the reports cut energy use by about 2%, an effect comparable to a short-term price increase of 11% to 20%. The heaviest users cut back the most, by over 6%. Nothing changed except what people could see.
So How Do We Handle the Surprises?
You won’t eliminate surprises. Even the best systems thinkers get surprised. But you can be surprised less often, and recover faster. Here’s the practice I’ve settled into:
Look for the pattern before reacting to the event. Assume there’s a feedback loop you haven’t spotted yet. Ask what’s on the other side of your boundary. Identify the current limiting factor, then ask what the next one will be. Multiply your time estimates. And when people behave badly, ask what they can see and what they’re rewarded for before deciding they’re the problem.
System Traps and Opportunities: Eight Patterns That Keep Smart People Stuck
Some systems aren’t just surprising. They’re perverse. Their structure reliably produces bad outcomes, no matter who’s inside. Systems thinkers call these recurring structures archetypes. I think of them as traps. But every trap, once you recognize it, is also an opportunity, because you can redesign your way out.
Trap 1: Policy Resistance (Fixes That Fail)
When several actors pull a system toward different goals, any push from one side triggers harder pushback from the others. Everyone works furiously to keep the system exactly where no one wants it.
The starkest example I know: in the 1960s, Romania’s government wanted a higher birth rate, so it banned abortion and contraception. Births tripled at first. Then families, who could not afford more children, resisted. The birth rate drifted back down through dangerous illegal abortions, maternal deaths rose sharply, and many unwanted children ended up in orphanages. When the regime fell, the first law the new government repealed was that ban.
Sweden, worried about the same issue in the 1930s, took the opposite approach. Instead of fighting families, it looked for a goal everyone shared: every child should be wanted and well cared for. It invested in health care, education, and family support. The birth rate has risen and fallen since then without panic, because the nation aligned around something deeper than a number.
The way out: stop overpowering resistance. Bring the actors together and find a larger goal they can all pull toward.
Trap 2: The Tragedy of the Commons (The Grand Banks Cod)
When a resource is shared, each user gets the full benefit of taking a little more, but the cost of overuse is spread across everyone. The feedback from the resource to the individual is weak or delayed. So everyone takes a little more, until the resource collapses.
For nearly five centuries, cod off Newfoundland seemed inexhaustible. Then came factory trawlers. Catches on the Grand Banks peaked at about 810,000 tons in 1968. Over the following decades, the spawning stock fell by roughly 93%. Making things worse, a 1989 internal review found that Canada’s fisheries department had overestimated cod populations by as much as 100% in the years before the collapse. Delayed, distorted information, the worst possible combination. On July 2, 1992, Canada declared a moratorium. About 30,000 people in Newfoundland and Labrador lost their jobs overnight, the largest layoff in Canadian history. The commercial moratorium wasn’t lifted until 2024, and the new quota was a small fraction of the pre-collapse catch.
The classic solutions are to educate and appeal to conscience, to privatize the resource so each user bears their own consequences, or to regulate access with “mutual coercion, mutually agreed upon,” like traffic lights and parking meters.
But there’s a fourth path I find inspiring. The political scientist Elinor Ostrom spent decades studying communities that managed shared resources well for generations, from Swiss alpine pastures to Philippine irrigation systems to Maine lobster fisheries. She showed that local people, given the right to set and enforce their own rules, often succeed where both markets and governments fail. In 2009 she became the first woman to win the Nobel Prize in economics.
The way out: restore the missing feedback, so that people who use the commons feel the consequences of overusing it.
Trap 3: Drift to Low Performance (Eroding Goals)
This one is quiet and deadly. Performance slips a little. Instead of fixing it, people adjust the standard to match. “That’s about what you’d expect.” “We’re not much worse than last year.” The lower standard allows the next slip. Round and round.
Think of the boiled frog: the water warms so slowly that the frog never feels the moment to jump.
On January 5, 2024, a door plug blew off an Alaska Airlines Boeing 737 MAX 9 a few minutes after takeoff from Portland. The NTSB found that four bolts meant to hold the plug in place had never been installed when the plane left the factory. Its final report placed responsibility on Boeing’s inadequate training, guidance, and oversight, and called for an independent review of the company’s safety culture. Missing bolts don’t happen because of one bad afternoon. It’s hard not to read them as the visible tip of standards that had eroded, step by step, over a long time.
The same drift happens in labs (QC thresholds loosened “just this once”), in personal health (a workout routine that slowly fades), and in finances (spending creep that becomes the new normal).
The way out: keep standards absolute. Even better, anchor them to your best past performance instead of your worst. The same reinforcing loop that drags you down can pull you up: the better things get, the harder you work to make them better.
Trap 4: Escalation (The Uber-Didi Price War)
When each side sets its goal by trying to beat the other side, you get a reinforcing loop of escalation: arms races, smear campaigns, ever-louder ads, and price wars.
When Uber entered China, it went head-to-head with local rival Didi, and both subsidized rides heavily to win market share. Each discount was met by a bigger one. Uber’s CEO admitted the company was losing more than $1 billion a year in China. In August 2016, Uber sold its China operations to Didi in exchange for a stake in the combined company, ending a war neither side could afford to keep fighting.
The way out: avoid getting in. If you’re already in, either refuse to compete (unilateral disarmament) or negotiate new rules that put a cap on the escalation.
Trap 5: Success to the Successful (The Matthew Effect)
When winners are rewarded with the means to win again, the gap between winners and losers widens until the losers are eliminated. Anyone who’s played Monopoly knows how this ends.
In 1968, sociologist Robert K. Merton gave this pattern a name academics instantly recognized: the Matthew effect, from the biblical line “to every one that hath shall be given.” Merton documented that eminent scientists tend to get disproportionate credit for work similar to that of lesser-known colleagues, and later research showed the same effect in funding and resources. More grants bring more data, more papers, more prestige, and more grants. Talented early-career researchers without that head start face a steeper hill.
Ecologists call the same structure competitive exclusion: two species competing for exactly the same niche can’t coexist, because whichever has the slightest edge gradually takes everything.
The way out: diversify into a different niche instead of competing head-on, cap how much of the pie any one winner can take (antitrust laws do this in theory), and periodically level the playing field. Many funding agencies, for example, have dedicated programs for early-career investigators for exactly this reason.
Trap 6: Shifting the Burden (Addiction)
A problem appears. Instead of fixing its root cause, an intervention relieves the symptom. It works, so it’s used again. Meanwhile the system’s own ability to solve the problem withers, so it needs more of the intervention. That’s addiction, and it doesn’t only involve drugs.
Antibiotics are my field’s sobering example. They are miraculous. But overuse and misuse let resistant bacteria flourish, which demands stronger antibiotics, which breeds more resistance. A landmark analysis published in The Lancet estimated that in 2019, drug-resistant bacterial infections directly caused about 1.27 million deaths, more than HIV or malaria, and were associated with nearly 5 million.
You’ll find the same structure in a business addicted to discounts to hit quarterly numbers, a farm addicted to pesticides after its natural predators are gone, or a team that relies on one heroic person to fix every crisis and never builds the skills itself.
The way out: if you’re the intervenor, help the system strengthen its own capacity, then step back. If you’re the dependent one, rebuild your own capability before withdrawing the crutch, and start now, because withdrawal only gets harder.
Trap 7: Rule Beating (The Hanoi Rat Hunt)
When rules are imposed, people get creative in following the letter while defeating the purpose.
In 1902, French colonial authorities in Hanoi were desperate to control a rat infestation as plague spread. They offered a bounty of one cent per rat tail. Tails poured in. Then officials began noticing tailless rats running around the city. People were catching rats, cutting off the tails, and releasing them to breed more rats with more tails. Some on the city’s outskirts reportedly began farming rats. The bounty was scrapped.
The way out: design rules so that creativity flows toward the rule’s purpose, not toward gaming it. And when you discover rule beating, treat it as feedback about the rule, not just about the cheaters.
Trap 8: Seeking the Wrong Goal (The Wells Fargo Scandal)
A system will faithfully deliver whatever goal you actually measure, even if it’s not what you meant.
Wells Fargo once aimed for each customer to hold eight financial products with the bank. Employees faced intense pressure to hit aggressive daily sales targets. The system delivered exactly what it was told to optimize: accounts. A review eventually flagged about 3.5 million potentially unauthorized accounts. In 2020, the bank agreed to pay $3 billion to settle criminal and civil investigations, and its former CEO was fined and barred from banking for life.
Academia has its own version: when you count papers, you get more papers, not necessarily more discovery.
The way out: define goals that reflect the real welfare of the system, and never confuse effort with result. Otherwise you’ll build a system that produces lots of effort and very little of what you wanted.
Where to Intervene in a System: The 12 Leverage Points, From Weakest to Strongest
So you’ve spotted a trap. Now what? Where do you push?
The most practical tool I took from this book is a ranked list of leverage points: places in a system where a small shift can produce a big change. Here’s the twist that makes it fascinating. The places where we spend most of our energy are usually the weakest levers. And when people do find the powerful levers intuitively, they often push them in the wrong direction.
Here’s the list, from least to most effective, with how I think about each one.
The Low-Leverage Levers: Numbers, Buffers, and Physical Structure
12. Numbers (parameters): tax rates, interest rates, budgets, salaries, thresholds. This is where 90% or more of our arguments happen. Parameters matter to the people standing in the flow, but they rarely change how a system behaves. Tweaking interest rates hasn’t eliminated business cycles. Tweaking a script’s p-value cutoff won’t fix a broken analysis design.
11. Buffers: the size of stabilizing stocks relative to their flows. Remember Toyota’s two-day supply of valves? After the 2011 earthquake and tsunami severed its supply chains, Toyota identified about 1,500 at-risk parts and built a business continuity plan requiring suppliers to stockpile two to six months’ worth of key chips. When the global chip shortage hit in early 2021, rivals suspended production while Toyota kept building cars for months. The same company that taught the world just-in-time learned where not to be lean. Buffers cost money, though, so they’re a trade-off, not a magic fix. (Even Toyota eventually had to cut production later in 2021 as the shortage dragged on.)
10. Stock-and-flow structures: the physical plumbing, like roads, factories, building layouts, and supply chains. Powerful, but usually slow and expensive to change. The leverage is in designing them right in the first place.
The Middle Levers: Delays and Feedback Loops
9. Delays: the lengths of time relative to how fast the system changes. Shortening an information delay can calm an oscillating system, as P&G found when it started sharing real sales data with suppliers. Many physical delays can’t be changed, so often the smarter move is to slow your own reaction to match them.
8. Balancing feedback loops: the strength of the system’s self-correction. Strong balancing loops, like a free press, transparent reporting, or a robust QC step in a pipeline, keep a system honest. When we remove them to save money (“we’ll skip the replicate this time”), we lose the system’s ability to correct itself.
7. Reinforcing feedback loops: how fast the snowballs roll. Slowing a runaway loop is usually more powerful than strengthening the balancing loop fighting against it. Progressive taxation, antitrust laws, and early-career funding programs all work by damping success-to-the-successful loops.
The High-Leverage Levers: Information, Rules, and Self-Organization
6. Information flows: who has access to what information. This is where I started getting excited. After the 1984 Bhopal chemical disaster, the U.S. passed a law in 1986 creating the Toxics Release Inventory, which simply required large manufacturers to publicly report their emissions of certain chemicals. There was no new emissions limit for those chemicals. Over the following years, reported releases of the original 300-plus listed chemicals fell by more than 45%. (Researchers still debate how much was disclosure itself versus other regulations, which is a fair caveat.) The Opower reports and the Dutch homes where families used less electricity simply because the meter was visible in the front hall make the same point: missing feedback is one of the most common causes of system failure, and restoring it is often cheap.
5. Rules: incentives, punishments, constraints. Rules define a system’s scope and boundaries. That’s why lobbyists crowd around lawmakers, and why the Hanoi bounty and the Wells Fargo targets went so wrong. If you want to understand a system’s deepest problems, look at its rules and at who has power over them.
4. Self-organization: the power to add, change, or evolve system structure. This is the system’s ability to invent its own solutions, like the Toyota suppliers improvising P-valve lines. Encouraging variety, experimentation, and diversity strengthens it. Rigid control and monoculture destroy it.
The Highest Levers: Goals, Paradigms, and Beyond
3. Goals: the purpose of the system. Change the goal and everything beneath it reorganizes. Wells Fargo’s goal of eight products per customer drove millions of fake accounts. A goal of “customers whose finances actually improve” would have driven entirely different behavior, from the same people, in the same branches.
2. Paradigms: the shared, unstated mindset from which goals, rules, and structures all arise. When Satya Nadella became CEO of Microsoft in 2014, the company’s culture was famously competitive and internally combative, and its stock had stalled. Inspired by psychologist Carol Dweck’s work on growth mindset, he pushed the company to move from being “know-it-alls” to being “learn-it-alls.” It sounds like a slogan. But it changed what got rewarded, how teams collaborated, and which bets the company was willing to make, including its big pivot to cloud computing. Over his first decade, Microsoft’s stock rose more than 1,000%, and its market value reached around $3 trillion.
Paradigm change is strange. It’s the hardest thing for a society to accept, yet in a single person it can happen in a moment, like scales falling from your eyes. You change paradigms by pointing persistently at the failures of the old one, and by speaking and acting confidently from the new one.
1. Transcending paradigms: the highest leverage of all is to hold no paradigm too tightly, including your own, and to realize that every worldview is a limited model of an immense reality. It sounds mystical, but I think it’s deeply practical. If no paradigm is the final truth, you’re free to choose the one that best serves your purpose, and to drop it when it stops working.
What the Leverage Points Taught Me About My Junior Scientists
Looking at that list, I finally understood why my junior scientists’ frustration persisted even after they’d learned better coding tricks. Better scripts are parameter-level changes, number 12, the weakest lever on the list.
The real lever was the paradigm. “Every dataset is a brand-new battle” is a worldview. The alternative is: “Every dataset is a variation on a structure I’ve seen before.” Once someone truly adopts that second view, everything downstream changes. They start writing modules instead of one-off scripts (hierarchy). They add checkpoints (balancing loops). They log and review results (information flows). They define what “done” means before they start (goals). The datasets don’t get any easier. The person gets stronger.
How to Build a World of Systems and Live in It
Here’s the most humbling part of the book, and the part I appreciated most.
People who discover systems thinking often make the same mistake. We assume we’ve finally found the key to prediction and control. Build a big enough model, map enough loops, and we’ll master the system.
It doesn’t work that way. Complex systems can’t be fully predicted or controlled. My junior scientists were right about that: you can’t predict the future with the past. But systems can be designed and redesigned. You can expect surprises, learn from them, and even profit from them. The metaphor I love most is that we can’t control systems, but we can dance with them. Like whitewater kayaking or playing music in a group, it requires staying wide awake, paying close attention, and responding to feedback.
There are about fifteen guidelines for living in a world of systems. Here are the ones I’ve found most useful, translated into practice for life, career, and business.
Get the Beat Before You Make a Move
Before you disturb a system, watch how it behaves. Learn its history. Find or make a graph of real data over time, because memories are unreliable.
It’s the same principle behind exploratory data analysis: explore before you model. Plot it. Look at distributions, batch effects, outliers. You’ll be amazed at how often the data tells a different story from what everyone “knows.” The same goes for joining a new team or entering a new market. Spend the first weeks observing, not fixing.
Expose Your Mental Models to the Light of Day
Draw your assumptions. Write them down. Share them so others can poke holes in them. A model that lives only in your head can’t be tested. This is exactly why reproducible, documented pipelines matter in science: they turn a private mental model into a public one others can check.
Honor, Respect, and Distribute Information
Most system malfunctions trace back to biased, late, or missing information. If you lead a team or run a business, the cheapest high-leverage move available to you is usually to make the right information visible to the right people at the right time. Dashboards that show what matters, honest postmortems, open books.
Pay Attention to What’s Important, Not Just What’s Quantifiable
We tend to measure what’s easy to count: papers, revenue, hours, followers. Quality, trust, curiosity, and wellbeing are harder to count, so they get ignored, and then they erode. Wells Fargo counted accounts. It didn’t count trust. Just because you can’t put a number on something doesn’t mean it isn’t one of the most important stocks in the system.
Make Feedback Policies for Feedback Systems
Static rules break in dynamic systems. Good policies adjust themselves based on the system’s state. A retirement plan that automatically rebalances, a budget that responds to actual spending, a pipeline that flags anomalies and halts before garbage propagates: these build learning into the system itself.
Locate Responsibility Within the System
Before blaming competitors, the economy, or the people below you, ask how the system’s own structure is producing the result. When a problem keeps recurring, the first question should be “What in our setup makes this mistake easy to make?” and only then “Who made it?”
Go for the Good of the Whole, and Expand Your Time Horizons
Optimizing one part often hurts the whole. And short-term thinking is how commons get destroyed and how careers stall. Ask what a decision does to the whole system, and what it looks like in five or ten years, not just this quarter.
Defy the Disciplines and Stay a Learner
Real problems don’t respect academic boundaries. Good pipeline design borrows heavily from software engineering and manufacturing. Some of the sharpest business insights come from ecology. And stay humble: when you’re wrong, and you will be, admit it, learn, and adjust. A system thinker who can’t say “I don’t know” stops learning, and a system that stops learning becomes brittle.
Celebrate Complexity, and Don’t Forget to Care
The world is messy, nonlinear, and full of surprises. That’s not a flaw to be engineered away. It’s the source of creativity, diversity, and life itself. Living well in a world of systems takes more than calculation. It takes intuition, compassion, vision, and a moral compass.
Putting It Together: A System for Life, Career, and Business
So here’s what I’d say now to those junior scientists, and to anyone who feels like every project, every month, every new challenge is another battle fought from scratch.
You don’t need one script for everything. You need structure:
- Your work: reusable modules (hierarchy), quality checkpoints (balancing loops), logs and reviews (information flows), a clear definition of done (goals), and a little slack in the schedule (buffers).
- Your career: skills are a stock. Fill it steadily and plug the drains, like the time you lose to work that teaches you nothing. Find a reinforcing loop where each project builds credibility that leads to better projects. And watch for the Matthew effect working against you, so you can choose a niche where your edge compounds.
- Your money: savings are a bathtub. Plug the leaks as seriously as you chase the income. Let compounding work over long horizons. And watch for the drift to low performance called lifestyle creep.
- Your business: design your flywheel, build buffers where shocks would be fatal, share information honestly, and set goals that reward the outcome you actually want.
If you’ve read my take on The Lean Startup, you’ll notice the same idea from a different angle: build-measure-learn is nothing more than a fast feedback loop designed to shorten the delay between an assumption and the truth. And in my piece on Entrepreneur Revolution, the “industrial age mindset” is really a paradigm, the highest and hardest leverage point of all.
Final Thoughts: You Can’t Control a System, but You Can Dance With It
I’ll be honest about the limits here. Systems thinking won’t tell you what will happen next year. It won’t hand you a formula. Its models are designed to explore “what if,” not to predict. And it’s easy to get lost in it, drawing so many loops and arrows that you lose the question you started with. Even the leverage-points ranking is a tentative, slippery list, not a recipe, and I appreciate that honesty.
But here’s what it did give me: a new way of seeing. Problems stopped looking like random events and started looking like the behavior of structures. I stopped asking “Who’s to blame?” and started asking “What’s the system?” That shift alone has made me a better scientist, a better teacher, and a better builder.
My junior scientists were right that every dataset is different. But they were wrong that every dataset has to be a battle. The Slinky’s bounce lived in its coils. The chaos in their work lived in its structure. Change the structure, and the behavior follows.
You can’t control the river. But you can learn to read the current, choose where to put your paddle, and enjoy the ride.
Now I’d love to hear from you: what’s one area of your life, career, or business where you keep fighting the same battle over and over? What feedback loop, delay, or trap do you think might be hiding underneath it? Drop it in the comments. I read and respond to every one.
If this changed how you see your work, subscribe to BullishBooks for more deep dives into the books that shape how we think, build, and grow.
