Why Perfect Designs Fail (And How to Intervene Where It Matters)

Magic? Maybe. More likely, there’s a critical gap in how the design engages with the living, behavioral, and institutional systems that determine its fate. This phenomenon is exactly what environmental scientist Donella Meadows has spent her career studying. Her framework for identifying leverage points in complex systems remains one of the most practical tools for creating lasting change. On a daily basis, it influences the way I personally approach not just engineering problems but design challenges, office dynamics, even parenting decisions.

Which “systems” are we talking about exactly?

Before I get into the list (and provide several relatable examples), I’d like to briefly discuss a few fundamental notions about systems.

The Word “System”

When you hear the word “system,” your first thoughts may center on engines, computers, assembly lines, or even the way you organize your closet. We tend to think of systems as procedures or complicated collections of parts – the result of human design and intent, used by humans as tools to reach a goal. While human beings become external actors to many systems, we’re just as often internal moving parts ourselves (and often not conscious of it).

More broadly speaking, a system is any set of interacting things and the manner in which they interact. Ant colonies. Traffic jams. The spread of forest fires. Your office dynamic. Your hot coffee and the room-temperature air around it. All are systems made up of interacting parts. There’s tremendous power in recognizing both parts of and patterns in their interactions. As you read on, I encourage you to stretch your mind and relate these concepts to systems you frequently interact with, or are affected by, but hadn’t previously considered systems.

Systems as Models

While systems can comprise real, physical, interacting parts, systems themselves aren’t real or physical. They’re an abstract way in which we organize our thoughts to understand and model behavior. By defining boundaries, parts, and interactions, we can create a model that helps us understand reality, but the model itself is not the reality. It’s an imperfect representation used for prediction and understanding. That’s not to say they aren’t useful. In many cases we can use them to predict behavior to extreme accuracy, but it’s important to recognize that they’re approximations.

“Closed” Systems

When we develop a model of a system, we’re examining an incomplete slice of a bigger picture. You can model your savings account as a simple closed system, for example, with a balance that either decreases (when you make a withdrawal) or increases (when you make a deposit or earn interest). Easy enough, and a decent approximation, but in reality, several additional influences external to your system model have measurable consequences: Interest rates are susceptible to federal policy changes. Inflation decreases purchasing power. Stolen login credentials or data breaches can result in a mysteriously empty account. By expanding the boundaries of our system model to include more parts and interactions, we can improve the accuracy of its approximation of reality, but ultimately, because there are no truly closed systems, there are no absolutely accurate models.

So far I’ve emphasized the limiting nature of closed systems, but there’s a beautiful flipside: You get to choose the boundaries based on what you’re trying to understand. It’s perfectly OK to start small and expand as needed. System models aren’t reality, but they are a tool, and you pick the right tool for the job.

All this is to say that system models aren’t divine truth. They blur, abstract, and assume, but even so, they’re frequently highly useful for predicting (or controlling) the emergent behavior of their parts. The following framework suggests how to recognize functional parts of systems, and when you need to modify a system, which modifications to those functional parts give you the most bang for your buck. Just as system models are incomplete and evolving, this framework isn’t handed down from on high, nor is it set in stone. It is, however, a generalized, helpful yet imperfect lens through which we can observe, analyze, and manipulate systems. Here begins Donella Meadows’ list of leverage points within a system, ranked in order of increasing effectiveness.

Donella Meadows’ Places to Intervene in a System

(Paraphrased and in order of increasing leverage)

12. Numerical parameters

11. The size of stocks, relative to flows

10. The structure of stocks and flows

9. The length of delays, relative to the rate of system change

8. The strength of self-correcting feedback loops

7. The strength of self-reinforcing feedback loops

6. The structure of information flow

5. Rules

4. Adaptability

3. Goals

2. Paradigms

1. Transcending paradigms

Those items are a bit vague on their own, so let’s dive right into it.

12. Numerical parameters

Imagine a city. This city has trees. They’re very nice trees. We want to devise a decision-making system to ensure they’re not going anywhere. We need to know what we’re working with, so we measure the city’s total current canopy coverage at 23%.

That’s nice to know, but that information isn’t really all that valuable without more context, so we measure it again, same time next year. Now we have 22.5% total canopy coverage. Compare that with last year, and we see that our canopy coverage has decreased by 0.5%. This number carries more weight than a single snapshot in time. We now have some vital information that a single snapshot in time couldn’t give us: The tree canopy coverage is decreasing.

Not only that – we’ve suddenly gained the power to predict the future. At its current rate, our city’s canopy coverage will drop to 12.5% in 20 years! That’s barely half of what it is now!

Let’s take annual measurements for the next 3 years and see how well we did.

21.8%. 20.8%. 19.1%

This additional context paints a much clearer and more worrisome picture. Each year we’re losing more than the year before. We have an acceleration. We can now see our rate of change itself changing, and just how quickly: At its current acceleration, our canopy coverage hits zero in only 4 more years!

Of course, this is an oversimplified example made to illustrate the point that the impact of a number depends entirely on its context. Is it …

• a value – what is the total canopy coverage now?

• a rate of change – how quickly is our canopy coverage changing, and in which direction?

• an acceleration – how quickly is our canopy coverage snowballing or stabilizing?

Some numbers are harder to change than others. In larger, complex systems where a number may depend on dozens of factors, this is increasingly true. Usually they can only be indirectly affected via changes made elsewhere in the system. The more we zoom in on individual parts, the easier numbers are to change.

The width in inches of a pathway. Irrigation spray head nozzle choice. The side slope ratio of a bioswale. When we aim to change a system, numerical parameters like these are often the first things we think to reach for. If they’re within our control, they’re usually relatively quick to change, but they’re also the least effective leverage point in this list. Not because they don’t affect things elsewhere in the system, but because they don’t affect the underlying structure or behavior.

11. The size of stocks, relative to flows

Savings accounts. Stormwater detention basins. Water-retaining soil amendments. In systems lingo, these are called “stocks,” and they function like shock absorbers. They’re the inventory, the reserves that add stability by buffering fluctuations. Reservoirs cushion the impact of droughts. Functional redundancy in a planting palette – planting multiple species that perform the same ecological function (e.g., nitrogen fixation) – acts as a buffer, so the system persists if one species fails. In short, the larger a stock, the more slowly its state responds to changes. Note the flipside: If a system needs to respond quickly to changes, smaller stocks may be more helpful.

These stock buffers can be relatively impactful, but in systems that lean toward the physical more than the abstract, they’re frequently bulky and difficult to change. Storage takes space.

10. The structure of stocks and flows

The layout of an irrigation system, as opposed to individual spray head selection. The effect on water runoff of a site’s topography and swale location, as opposed to slope ratios or swale depth. This leverage point represents the layout of the physical infrastructure, the plumbing that connects stocks and flows. Consider the impact on wildlife-vehicle collision rates with the addition of a wildlife overpass over a highway compared with the impact of simply lowering the speed limit.

Rearranging the network of stocks and flows itself can have a significantly greater effect than tweaking the knobs of an existing structure. In considering this leverage point, you may ask, “How do the stocks and flows interact in this network? Where are the buffers, and where are the bottlenecks? Where are the missing connections?”

9. The length of delays, relative to the rate of system change

How long does it take for a plan to be approved relative to the rate at which site conditions and client expectations evolve? What’s the delay between the time at which a culvert is clogged with sediment and the time at which a maintenance response occurs? How long after the introduction of a new social program will its effects manifest? Most systems don’t propagate changes immediately. How long a system takes to begin to react can have major consequences on the way it adjusts to changes.

If information propagates too slowly relative to how fast the system changes, problems emerge. The system state can overshoot its target, oscillating above and below target. As Meadows points out, anyone who has ever taken a shower on the other end of the house from the water heater can relate to this one on a personal level. (Too hot, better turn it down … too cold, better turn it up. Ouch!).

More often than not, delays are limitations that we have to work around. We can’t do much to change how quickly plants establish roots and mature. Construction doesn’t happen instantly. Clients don’t always pick up the phone. In these cases, it’s usually more effective to manage information flows accordingly, or slow down the rate at which things change, to better sync up with the delay length. Play the delay. Don’t let it play you.

8. The strength of self-correcting feedback loops

Feedback loops are a system’s way of automatically responding to changes in its state to maintain a target state. The actual, measured state (or really the discrepancy between the measured state and the target state) is “fed back” into the input. Just like that, we have a “smart” system that can react to changing conditions.

There are two types of feedback loops: Negative and positive. A fairly common misconception holds that negative feedback is bad and positive feedback is good. In fact, “positive” and “negative” here have nothing to do with value and everything to do with function. Both types of feedback “push on” the system state, and both self-adjust. The difference lies in the direction of the push and its strength over time:

• Negative feedback loops are self-correcting: The further the measured state is from the target state, the harder they push toward the target state. They balance, moderate, and stabilize. Negative feedback makes big differences smaller.

• Positive feedback loops are self-reinforcing: They push away from a starting point with increasing strength, off to infinity, if left to their own devices. They destabilize, amplify, and drive growth (or collapse). Positive feedback makes small differences bigger.

Of the two, the next stop on our list of leverage points is the negative (corrective) feedback loop. Why positive feedback loops are considered higher leverage will be apparent in a minute. For now, consider a smart irrigation controller that auto-adjusts runtime schedules based on the current soil conditions.

In this case, the state we’re concerned with is the soil moisture. A soil sensor takes a measurement of the current soil moisture and feeds that information back into the smart controller. The smart controller notes the difference between the measured soil moisture and the target soil moisture. If that difference is large (the soil is dry), the irrigation controller pushes the system strongly toward the target soil moisture (water for the full scheduled runtime). If that difference is low (the soil is already relatively moist), the irrigation controller only gently pushes the system towards the target soil moisture (water for only a fraction of the normal runtime). If the soil is already at its target moisture, the difference between measurement and target is zero, and there’s no push at all. (Skip this watering cycle). The result of this single negative feedback loop is a system that self-corrects to maintain a target value.

Negative feedback loops like these are everywhere. Your body shivers to warm up and sweats to cool down. Drivers merging onto a highway let off the gas as they approach the speed of traffic. The heater in your home turns off once room temperature reaches its target. These are even abundant in more abstract systems of human interaction. Your design may go through a series of revisions to approach an ideal outcome. Case studies of past projects may lead you to reform your design methodology over time.

Systems with negative feedback loops balance themselves out when things begin to drift too far in one direction. The stronger the negative feedback, the greater the resistance to change.

7. The strength of self-reinforcing feedback loops

Resistance to change can be helpful, but if we want to simultaneously drive change and resist stagnancy, positive feedback is needed as well. Without it, plants wouldn’t grow to face the light. Life wouldn’t evolve. Reforestation efforts would fall flat.

These are powerful forces – great to have on your side, and terrifying to have working against you. Recall that positive feedback drives growth or collapse by amplifying small changes and leaning into forward momentum. Depending on the context, this momentum may be undesired. The self-reinforcing feedback loop is the reason for both stock market booms and crashes. It’s the reason epidemics spread so quickly. It’s the reason that a decline in native flowers results in a decline in native pollinators, which results in a decline in native flowers, which results in …

Any time you hear of a “downward spiral” or “vicious cycle,” it’s most likely a positive feedback loop. These appear relatively high up on this list because, once the snowball gets rolling, it can be incredibly hard to stop. In general, it’s more effective to limit the strength of a runaway positive feedback loop than it is to strengthen a weak negative feedback loop.

6. The structure of information flow

Just as there’s more leverage in modifying the structure of stocks and flows than there is in modifying their magnitude, there’s more leverage in modifying the structure of feedback loops than in modifying their strength. In larger, more complex systems involving humans as moving parts, the real question here is, “Who has access to which information?”

Does the community have input on park design, or do only the city officials? Do property managers have access to real-time water usage data, or only monthly bills? Do the users of a site know where recycled water or native plants are used? How might a community’s water quality be affected if public consumer confidence reports weren’t required?

Meadows provided a great example:

'The Toxic Release Inventory – the U.S. government’s requirement, instituted in 1986, that every factory releasing hazardous air pollutants report those emissions publicly every year – suddenly informed every community of precisely what was coming out of the smokestacks in town. There was no law against those emissions, no fines, no determination of “safe” levels – just information. Still, emissions dropped 40 percent between then and 1990. They’ve continued to decrease since, not so much because of citizen outrage as because of corporate shame. One chemical company that found itself on the Top Ten Polluters list reduced its emissions by 90 percent, just to “get off that list.”'

Significant change can arise from simply passing information from point A to point B. Often the effect trickles down and influences everything lower on this list.

5. Rules

Accessibility requirements, municipal stormwater regulations, plant palette restrictions … the rules of a system constrain all structures and parameters discussed above. Changing the rules that govern a system can be rather difficult, but doing so can trigger massive change in its structure. There’s a reason so much money goes into lobbying. Those who make the rules wield enormous leverage over the system.

Systems develop within rulesets. Change your system, and you may play a better game. Change the rules, and you’ve got a different game entirely.

4. Adaptability

This leverage point refers to the rigidity of a system’s structure, its ability to “self-organize,” to evolve as its environment, rules, and goals change. Consider a landscape design firm as a system. How malleable is its business model? The internal operations? How much weight do employee concerns and suggestions carry? When demand arises for new design styles or new project types, can the firm offer them? When new regulations or technologies appear, can the firm adapt? Drafting by hand was the norm until CAD came onto the scene. Especially with the development of increasingly powerful design and automation tools, firms less willing or less able to adapt have faced increasingly powerful competition.

The same goes for landscape designs: As climate patterns shift and “hundred-year storms” arrive every decade, adaptive designs will continue to function while rigid designs fail. You may ask, How resilient will this design be in the face of a natural disaster? How does this design impact the resilience and adaptability of its surrounding ecosystem, environment, and community?

Whether or not we want it to, the world changes. If a system can’t change with it, its effectiveness will only decrease.

3. Goals

Continuing with the perspective of a landscape design firm as a system, even more influential than its adaptability are its goals. How does the firm define healthy internal operation? How does it define success? Project scale? Number of awards? Environmental impact? Changing the firm’s mission statement from “create beautiful spaces” to “restore ecosystem function” or “maximize sustainability” reshapes every decision.

Whether by natural processes or by design, everything further down in this list is a means to achieve a goal. Change the goal of a system, and as long as it’s sufficiently adaptable, its entire structure, including its rules, feedback loops, stocks, and flows, will change as necessary to meet that goal.

The next time you open a plan, consider the project’s goals and how they impact your design and decision-making process. Consider your personal goals as well. How aligned are your habits with those goals? Are some adjustments in order? At which leverage point in your system?

2. Paradigms

Next up are the paradigms or mind-sets from which goals form and methodologies develop. These are our deeply rooted beliefs, assumptions, and social agreements, usually so fundamental that they go unspoken. Human beings are part of the global ecosystem. Growth is a good thing. Land has monetary value. Land can be owned.

Even if some of these seem self-evident, they’re human constructs and perspectives that are just as susceptible to change over time as anything else. When we zoom back in on landscape architecture, we find a plethora of paradigms we take for granted, some of which aren’t entirely mutually compatible.

Change “landscape as status” to “landscape as stewardship,” and design goals may shift from manicured lawns to ecologically and environmentally sustainable landscapes.

Change “water as a commodity” to “water as a commons,” and design goals surrounding water use, quality, retention, and runoff may evolve quite a bit.

Change “human vs. nature” to “human as nature,” and designs may begin to draw (or blur) the lines between natural and manufactured spaces in a much different way.

Because they’re so ingrained and elusive, paradigms are difficult to change, let alone identify. Meadows suggests that the process of modeling a system “takes you outside the system and forces you to see it whole.” I’ve personally found this theory true to an extent, but I’ve found more success in identifying my own paradigms by interacting with other human beings with different backgrounds, cultures, and experiences.

1. Transcending paradigms

Meadows’ highest-leverage place to intervene is perhaps more like turning the whole idea of system models and paradigms on its head: No one paradigm is inherently more true than any other, and all paradigms are, at their core, temporary, imperfect notions about an evolving reality.

Bear with me just a little bit longer here if this is getting too woo-woo for you. The way I see it, the point is less “nothing matters,” and more “everything is considerable.” Put in the effort to recognize the paradigms you hold close, and remain flexible and open to new ones. Shifts in perspective can be revolutionary – more so than anything else on this list – but only if we give them the opportunity to happen in the first place. The capacity to question our assumptions and see the system, and our role in it, from a higher vantage point, is the ultimate leverage.

Just like that, we’ve reached the summit (and boy is it theoretical up here). “Transcending paradigms” sits at the peak of the list, but its true value is realized only when applied to the messy, complex reality wherein we work every day.

Back down to earth

Thinking in systems matters more than ever for landscape architects. You’re directly intervening in living systems, in watersheds, in communities, in ecosystems with their own momentum and feedback loops. With so many moving parts, and with pressure from several directions, the temptation for immediate, concrete control is understandable. But zoom in too far, fixate on plant spacing or paver specs, and you risk optimizing a single parameter while the larger system goals drift out of alignment.

Next time you're stuck on a problem, whether it's a design challenge, an office dynamic, or even a personal issue, run through this list. Ask yourself: Am I adjusting parameters when I should be restructuring information flows? Am I strengthening feedback loops when I should be questioning goals? The highest leverage often lies not in working harder within your current approach but in stepping back and intervening at a deeper level of the system itself. I hope this perspective brings as much benefit to you as it has to me over the years. Sometimes it really does feel magic.