Feedback Loops: How to Control Complex Systems

Complex systems appear to be very… complex. It’s like having a control panel with a thousand buttons in front of you; you have no idea what any of them do, or even where to look.

Most people spend their lives serving one or two buttons on the panel, rarely considering how the entire system operates. When your boss tells you that your job is crucial and absolutely cannot tolerate any sloppiness, you nod in agreement, thinking: If even my job is so important, those who control the big picture must be gods, right?

They are not gods. In fact, this system is not as hard to understand as you might think. This article will help you quickly grasp the essence of a system.

The core idea is that in any complex system—whether it’s a company, a nuclear power plant, or even a country—there are usually only a few causal relationships that are truly decisive.

Grasp these few, and you can understand it, control it, and predict its fate.

These critical causal relationships are what we call “feedback loops.”

What is a Feedback Loop? #

Whether you’re facing a large system or trying to understand one of its subsystems, don’t view it as a static machine. Instead, you should imagine it as a “living” organism with “inputs” and “outputs.” Take a factory, for example: it inputs raw materials and outputs products. But raw materials are purchased with money, and products are sold for cash, so, in essence, the factory inputs money and outputs money. The difference between the input and output, the portion that ultimately remains in the system, is the money it earns. Isn’t that simple to understand?

Next, from input to output, there’s a series of causal relationships: A leads to B, B leads to C… Actions thus produce consequences that propagate outward.

A feedback loop is when a causal relationship completes a circle and returns to its starting point: A leads to B, B leads to C, and C, in turn, acts upon A, forming a closed loop.

For instance, if you lower the price of a product (output), sales will increase; increased sales will dilute the unit fixed cost (input); diluted costs will allow you to further reduce the product price. This is a “positive feedback” loop, meaning the feedback strengthens your initial action, thereby amplifying the overall change.

There are many positive feedback phenomena in life:

• A child performs well, receives encouragement → encouragement boosts confidence → confidence, in turn, makes them perform better. The more a stock rises, the more people buy → the more people buy, the more the stock rises.

However, please note that the meaning of positive feedback does not necessarily equate to “positive growth.” A stock market crash is also a type of positive feedback: panic triggers selling, and selling creates even greater panic—this is essentially amplifying the emotion of “panic.” Positive feedback is an amplifying force, regardless of whether it amplifies positive or negative things.

True “negative feedback” refers to feedback that pulls the initial action in the opposite direction. For example, with your home air conditioner, once the room temperature rises above the set value, it starts cooling; when the temperature drops back near the set value, it stops. The initial action is cooling, and the feedback action is to stop cooling. Negative feedback is a force that corrects deviations and seeks balance.

Because the words “positive” and “negative” can easily evoke a sense of good versus bad, system dynamicists simply renamed them: positive feedback is called a “reinforcing loop,” and negative feedback is called a “balancing loop” [1].

Reinforcing loops bring growth—whether good growth or bad growth: compound interest, virus spread, snowballing debt, and so on, all fall into this category. Balancing loops, on the other hand, bring stability: thermostats, supermarket restocking, household budget control, and body temperature regulation are all typical examples.

The Chinese often say, “The way of heaven is to take from those who have too much and give to those who have too little”—this is essentially a balancing loop. “The way of man is to take from those who have too little and give to those who have too much”—this is a reinforcing loop. There’s a saying in the Bible, in the Gospel of Matthew: “For to everyone who has, more will be given, and he will have abundance,” which expresses the same principle, so much so that today, whenever we mention positive feedback mechanisms, we associate them with the “Matthew effect.”

Reinforcing loops and balancing loops are the two major keys to understanding and controlling systems. To effectively control a system, you just need to grasp these two core principles: if you want it to grow, you must activate its reinforcing loops; if you want it to be stable, you must establish negative feedback (balancing loops).

Practical Applications of Feedback Loops #

Feedback loops allow us to understand many things.

For example, why is losing weight so difficult? Because your body has a built-in balancing loop that strives to defend your current weight: once you reduce your intake, it lowers your metabolism and amplifies your appetite, doing everything possible to compensate for the calorie deficit [2]. You are not fighting fat; you are fighting negative feedback.

So, how does one get fat in the first place? There’s often a common reinforcing loop: eat more, fat increases → the heavier the body, the more strenuous activity becomes → thus, people move less → the less one moves, the greater the calorie surplus, and fat continues to increase. At this point, fat is no longer just the result of getting fatter; it becomes the cause of the next round of weight gain.

If you don’t want certain things to get out of control, you must convert positive feedback into negative feedback. For example, in an argument, a harsh word will invite an even harsher one, causing the conflict to amplify itself with unimaginable consequences. Therefore, experienced couples establish mechanisms to automatically trigger “cooling-down” actions when signs of escalating conflict appear: as soon as someone raises their voice, both parties pause; the more intense the emotions, the less they speak.

A more complex scenario is when a situation is controlled by more than one feedback loop. For instance, if your company’s project is behind schedule and the boss requires everyone to work overtime, the schedule catches up. “Slow progress → overtime → progress caught up” is a very effective balancing loop, but it’s a “fast” loop. However, there’s also a “slow” loop for improving progress:

• “Slow progress → improve processes, train employees, advance automation → faster progress”—this is the more fundamental solution! If overtime continuously consumes resources originally intended for review, training, and process improvement, the “fast” loop will squeeze out the “slow” loop.

In 2001, two management scientists from MIT’s Sloan School of Management, Nelson Repenning and John Sterman, studied process improvement in companies and discovered a common trap: the greater the pressure, the more the “work harder” “fast” loop squeezed out the “work smarter” “slow” loop [3].

Such organizations become increasingly less “smart,” exposing more and more problems, with fires breaking out everywhere. As a result, they become increasingly reliant on “firefighting,” essentially entering a vicious positive feedback cycle, eventually turning the company into a forest that can only be barely maintained by constant firefighting.

Similarly, emotional weight loss is also an embodiment of the “fast” loop squeezing out the “slow” loop. Today, you fast if you gain a pound; tomorrow, you reward yourself if you lose half a pound; every day is spent correcting numbers without establishing long-term eating and exercise habits… The result is that the busier the “fast” loop becomes, the harder it is for the “slow” loop to initiate and function.

This is the true big picture perspective. With a feedback loop perspective, you can accomplish two major things.

Closed-Loop Control: The Key to System Regulation #

The second major thing you can accomplish is “closed-loop control” [4].

Its principle is very simple. Imagine you’re given a gun you’ve never used before and asked to aim at a target one hundred meters away. You don’t understand ballistics, you don’t know the wind speed or direction, and the gun’s sights might even be crooked. What should you do? The answer is: take one shot first.

After firing, observe the point of impact: if it’s too far left, aim a bit to the right for your next shot; if it’s too high, aim a bit lower. By making small adjustments each time, you can hit the bullseye within a few shots. You’ve used a balancing loop to find the bullseye. You don’t need any detailed knowledge; you just need effective feedback.

Governing a country is actually quite similar, for example, China’s reform and opening up. There was no precedent for how a large nation could transition to develop a market economy. Deng Xiaoping’s wisdom lay in employing closed-loop control, exploring forward by “taking a shot, observing feedback.” Today, a policy is introduced; tomorrow, a special economic zone is established; the day after, the stock market is opened. If the effect is good, it’s continued; if the effect is not good, it’s adjusted promptly. This is like driving a car: as long as you receive effective feedback, you can freely control the steering wheel without being fixated on every detail of economic principles.

Laozi’s saying, “Governing a great nation is like cooking a small fish,” shares a similar subtlety: governing a country is not an impossibly complex task; controlling its complexity is akin to cooking a delicate dish. Of course, a more important layer of meaning in this saying is that a nation’s economy is as fragile as a small fish and should not be tossed around carelessly—here, “slow” loops are needed more than “fast” loops—governing a great nation should not be as crude as making pancakes.

Opposite to closed-loop control is “open-loop control”: it requires various calculations beforehand, predicting the consequences of actions, and then issuing and executing commands with optimal force—but without making any fine-tuning based on feedback results. Having understood closed-loop control, you might find this open-loop control very strange: with such a complex system, how can one make such precise calculations?

However, a large number of things in life are precisely manifestations of open-loop control. For example, when certain departments issue policies, consultants submit proposals, or brands create advertisements, no feedback mechanisms are ever considered. This even includes how most hospitals currently treat patients: typically, medication is prescribed, the patient is discharged, and then they go home—very few doctors follow up months later to find out if the patient has recovered, has transferred to another hospital, or is in a worse situation. Don’t you find it absurd that such important work lacks a formal feedback mechanism?

Open-loop relies on sagacity, closed-loop relies on correction—with feedback, you don’t have to be omniscient.

Cold Start: Igniting the Growth Flywheel #

The second major thing you can accomplish is “cold start” [5].

Any endeavor that aims to grow and expand needs to go through a positive feedback process, which entrepreneurs often call the “growth flywheel.” However, positive feedback can only amplify “what is there,” it cannot create “something” out of “nothing.” For example, if you want to become famous through writing, you need people to share your articles—but you have no readers, so no one shares → no one shares, and you have even fewer readers. So, how can this flywheel be made to spin?

“Cold start” is about providing the initial “from 0 to 1” impetus: whether it’s asking industry leaders for recommendations or investing capital in advertising, you need to create the first wave of “out of nothing” seed users or resources.

The launch mechanism is different from the operating mechanism. A mature company will operate largely automatically: users spontaneously spread its products, and various merchants will automatically seek cooperation—but its initial launch phase can be entirely manual, inefficient, subsidized, or even involve the founder personally acquiring users one by one.

The core principles of “cold start” can be roughly summarized into four points:

• First, do not plan the starting point with the end in mind. Users, fame, and capital are products of the large loop operating fully, not the driving fuel for the first round. The true output of the first round is a viable prototype, a presentable result or work. Use your work to gain proof, and use proof to gain credibility.
• Second, the launch process might not be glamorous, but the results must be real. The first batch of users cannot be fake traffic numbers; they must be real people genuinely interested in you. Even if the first success is small, it must leave you with evidence, reusable methods, and valuable accumulated credibility. Otherwise, even the biggest buzz is just putting makeup on a dead system.
• Third, concentrate your firepower. Ten seed users scattered across the country will only be ten isolated points; only by concentrating them into a specific department, a target community, or a critical moment can a true breakthrough point be formed.
• Fourth, the output of the first round must be able to drive the second round. One delivery should leave a replicable case study; one service should leave a strong recommendation; one transaction should leave data, relationships, or cash. Otherwise, if each round requires re-investing capital and effort to acquire users, that’s not a “flywheel” but more like “artificial respiration.”

These four principles can be fully understood through a case study. In 2014, Tencent sought to promote WeChat Pay, but its biggest obstacle was the lack of user motivation to link bank cards—Alipay had already been deeply cultivating this for ten years, so how could WeChat quickly catch up? WeChat’s “cold start” strategy cleverly leveraged the custom of “giving out red envelopes during Chinese New Year.”

It did not take “promoting payments” as the end goal for its starting point; instead, it concentrated its firepower on the socially dense moment of Chinese New Year. As a result, from New Year’s Eve to the afternoon of New Year’s Day, over five million people participated in the red envelope grabbing activity. To withdraw the money they grabbed, they had to link a bank card. The red envelopes were gone, but the bank cards remained: in the social loop, the interpersonal relationships of Chinese New Year were the input, and the act of millions grabbing red envelopes was the output; then, in the payment loop, these users who had completed card linking became the first batch of inputs for the payment loop… Thus, a complete closed loop was formed. Jack Ma was “stunned” by this move, later calling Tencent’s operation a “Pearl Harbor attack” [6].

In fact, the key to Tencent’s “cold start” being so effective was that WeChat itself is a social tool, which was its existing first-mover advantage in social networking. And even so, it still had to “ignite” with real money. WeChat Pay truly reached a national scale only after the “shake-to-win” activity during the CCTV Spring Festival Gala the following year [7]. At that time, Tencent spent 53.03 million yuan just to acquire the exclusive new media interaction rights for the CCTV Spring Festival Gala, and the cash red envelope prize pool for “shake-to-win” exceeded 500 million yuan, provided by multiple corporate sponsors [8], all to prompt national viewers to shake out 120 million red envelopes in just ten minutes.

“Cold start” is costly, involves huge investment, and doesn’t guarantee success. Therefore, people always ask: Is it worth it?

First-Mover Advantage and Late-Mover Disadvantage #

Why should we invest our own resources into a “cold start” to gain a first-mover advantage? Why not wait for others to complete the “from 0 to 1” breakthrough, and then we directly pursue “from 1 to N” expansion?

For example, in the 1990s, Motorola launched an ambitious “Iridium project,” spending over five billion US dollars to build a global satellite phone network… The result was that terminal devices were too bulky, and call charges were too high, making it unaffordable for users, ultimately leading to bankruptcy protection in 1999 [9]. Given this, why shouldn’t we wait for someone else to prove the feasibility of a path, and then act with lower costs and higher efficiency to execute the plan and capture their market?

Many believe that the development of China’s manufacturing industry over the past four decades is a successful case of the so-called “late-mover advantage”: industries such as home appliances, high-speed rail, mobile phones, photovoltaics, and electric vehicles were all achieved through introduction, digestion, absorption, and re-innovation—a model of “others pave the way, we accelerate.” Chinese economist Justin Yifu Lin once sang high praises for the “late-mover advantage”—however, economist Professor Xiaokai Yang specifically warned that late-movers should not only see the advantages but also be wary of the “late-mover disadvantage” [10].

The reasoning is as follows: first-movers undertaking a “cold start” indeed face significant risks and encounter various difficulties, having to solve complex problems first—but in the process of solving these problems, they accumulate valuable experience and capabilities—such as institutional innovation, optimizing the investment environment, and fostering an original spirit—these accumulations enable them to solve new problems and thus potentially maintain a continuous leading position. While “copying homework” from behind certainly has lower costs, if you haven’t solved those problems independently, you won’t learn the ability to solve them, and you will only be able to continue following behind…

From “cold start” → first-mover advantage → ability to undertake the next “cold start,” this itself is also a reinforcing loop.

Perhaps you might say that first-movers indeed often “die on the beach”—didn’t we mention this when discussing “opportunity windows”? The timing for entry should be more appropriate after the dominant category emerges. That’s right, but true first-mover advantage isn’t about being the first to rush into uncharted territory; it’s about being the first to close your own loop when a path has just been proven and market positions haven’t been fully occupied. You don’t have to be the first to make mistakes, but you absolutely cannot wait until others have converted standards, users, talent, and channels into their own existing stock before entering the market.

Why can “Made in China” products occupy such a vast market share yet yield extremely low profits? Because lacking a first-mover advantage means lacking pricing power; you can only survive by externally lowering prices and internally squeezing costs.

How can one stand idly by and not strive for such a great opportunity? First-mover advantage is the true advantage; being a late-mover is merely a reluctant choice when there are no other options.

Positive Feedback and Negative Feedback: Drive and Balance #

Here, I particularly want to say a few words in defense of positive feedback. Positive feedback indeed possesses instability, can go out of control, and signals danger and trouble, always being associated with negative terms like bubbles and bank runs. In contrast, negative feedback is an important regulatory mechanism; it is stable, rational, understands correction, and is favored by managers and advocated by those who practice health preservation.

But only positive feedback can bring true growth. Positive feedback is the mechanism by which things emerge and evolve. Relying solely on regulation cannot create new things. Only after a reinforcing loop exists is it worthwhile to build a balancing loop for management.

It’s not to say we shouldn’t prioritize health, but if a person wants to flourish, they must possess at least one reinforcing loop: always have the next thing they want to do, and let the thing already completed become the driving fuel for the next.

To paraphrase a famous saying: Lose negative feedback, lose a lot; lose positive feedback, lose everything.

You need to “live,” not just “be alive.”

Notes

[1] Sterman, John D. Business Dynamics: Systems Thinking and Modeling for a Complex World. Boston: Irwin/McGraw-Hill, 2000.

[2] Hall, Kevin D., and Juen Guo. “Obesity Energetics: Body Weight Regulation and the Effects of Diet Composition.” Gastroenterology 152, no. 7 (2017): 1718–1727. https://doi.org/10.1053/j.gastro.2017.01.052.

[3] Repenning, Nelson P., and John D. Sterman. “Nobody Ever Gets Credit for Fixing Problems That Never Happened: Creating and Sustaining Process Improvement.” California Management Review 43, no. 4 (2001): 64–88. https://doi.org/10.2307/41166101.

[4] Åström, Karl J., and Richard M. Murray. Feedback Systems: An Introduction for Scientists and Engineers. 2nd ed. Princeton, NJ: Princeton University Press, 2021. https://www.cds.caltech.edu/~murray/FBS/Second_Edition.html.

[5] Chen, Andrew. The Cold Start Problem: How to Start and Scale Network Effects. New York: Harper Business, 2021. https://www.harpercollins.com/products/the-cold-start-problem-andrew-chen.

[6] 《微信红包“走红”，引爆 2014 移动支付年》，搜狐财经，2014 年 2 月 25 日。http://business.sohu.com/20140225/n395599166.shtml.

[7] 《春节红包战：他们到底在争什么？》，新浪科技，2021 年 2 月 12 日。https://finance.sina.com.cn/tech/2021-02-12/doc-ikftssap5433270.shtml.

[8] 《从钟表自行车到互联网大厂：春晚背后的顶级流量大战》，人民网，2020 年 1 月 16 日。https://media.people.com.cn/n1/2020/0116/c40606-31550446.html；《除夕上演红包决战，春晚“摇一摇”达 110 亿次》，《人民日报》/人民网，2015 年 2 月 20 日。https://finance.people.com.cn/money/n/2015/0220/c42877-26584574.html.

[9] “Iridium, Bankrupt, Is Planning a Fiery Ending for Its 88 Satellites.” The New York Times, April 11, 2000；“The Rise and Fall and Rise of Iridium.” Air & Space / Smithsonian Magazine. https://www.smithsonianmag.com/air-space-magazine/the-rise-and-fall-and-rise-of-iridium-5615034/.

[10] Yang, Xiaokai. “Late-Mover Disadvantage” (speech at Unirule Institute of Economics in December 2000).