Why Systems Change Fails When It Tries to Prove Itself Too Early

Jun 12, 2026

10 min read

1 views

61%

The hardest problem in change is not action, it is coordination

Most large social problems are not solved by a better idea in a lab, a stronger grant proposal, or a heroic organization with perfect execution. They persist because they are network problems: many actors, each making locally rational choices, collectively produce a system that no single actor can redesign alone. That is why so many well funded initiatives feel busy yet strangely ineffective. They create programs, services, pilots, and metrics, but they do not shift the relationships, incentives, and shared sensemaking that actually govern behavior.

This creates a deeper question than “what intervention works?” The real question is: how do you finance and learn for change when the thing you are trying to change is a living network, not a machine?

That question matters because systems do not respond well to the logic of certainty. They respond to patterns of trust, coordination, feedback, and adaptation. If you fund only what can be neatly attributed, you often starve the very conditions that make durable change possible. If you insist on proof before exploration, you can end up rewarding the parts of a system that are easiest to measure rather than the parts that matter most.

In complex systems, the first bottleneck is rarely knowledge. It is the ability to organize learning across people who do not naturally coordinate.

Why networks are not programs, and why funding keeps confusing the two

A program is built to deliver a defined output. A network is built to enable many actors to align, adapt, and amplify one another. The difference sounds subtle until you try to fund real change. Program logic asks for a bounded intervention, a clear owner, a timeline, and discrete outcomes. Network logic asks for convening, translation, trust building, shared language, and the slow formation of a field that can act together.

This is where traditional funding often misfires. It tends to treat collaboration as overhead, relationships as intangible, and experimentation as a luxury. But in systems change work, those are not extras. They are the infrastructure. If the system is fragmented, the most valuable asset may be the one thing that cannot easily be counted: the ability of diverse actors to recognize one another as part of the same problem and the same solution space.

Consider public health. Reducing childhood obesity, opioid deaths, or vaccine hesitancy is not just about having the right medical intervention. It requires schools, clinics, community groups, policymakers, families, insurers, local media, and often employers to move in complementary ways. A hospital can treat patients one by one, but it cannot by itself reshape the food environment, social norms, access pathways, or political narratives that determine population outcomes.

A common mistake is to ask public health to behave like manufacturing. But the unit of change is not a product moving down a line. It is a web of decisions. If the problem is distributed, the solution must be distributed too.

The hidden tension: evidence wants closure, systems want adaptation

Here is the core tension connecting change networks and realist public health thinking: evidence systems are often designed to close a question, while complex systems require us to keep questions open long enough to learn what actually works, for whom, under what conditions, and why.

That may sound like a technical distinction, but it has huge practical consequences. A simple evaluation model assumes that an intervention has a stable effect that can be isolated and measured. Yet in real-world systems, the same action can produce different outcomes depending on local context, timing, culture, power, and existing relationships. A school nutrition initiative may thrive in one district because parent champions are active and administrators are aligned, while failing in another because the community sees it as top-down intrusion.

This is where realist synthesis becomes so useful. Instead of asking only whether something works, it asks what mechanisms are activated in what contexts to produce what outcomes. That shift matters because it respects complexity without surrendering rigor. It says that success is not a universal property of an intervention. It is a relationship between mechanism and context.

Now connect this to funding. If money is allocated only after certainty is established, then financing arrives too late for the learning that generates certainty in the first place. If funding requires tidy attribution before a network has even formed, it can force actors to simplify, isolate, and compete. In effect, the funding model may destroy the conditions under which system change becomes possible.

The paradox is brutal: the more complex the system, the more the work depends on uncertainty, and the more likely conventional financing is to punish uncertainty.

A better model: fund the learning system, not just the intervention

The most useful way to think about system change is to stop viewing finance as a reward for finished solutions and start viewing it as a way to build a learning system. A learning system does three things at once: it connects people, tests hypotheses, and moves resources based on what is emerging rather than what was predicted in advance.

This changes the role of funders, intermediaries, and practitioners. They are no longer just buyers and deliverers of services. They become participants in a feedback architecture. The question shifts from “Did this project hit target X?” to “What did this network learn, who changed behavior, and what new capacity was created to keep adapting?”

Think of a city trying to reduce homelessness. A conventional approach might fund shelters, case management, or housing subsidies separately and evaluate each in isolation. A network approach would ask something different: where are the coordination failures? Are outreach teams, landlords, health providers, and benefit systems sharing information? Are there bottlenecks in placement? Are there feedback loops that cause people to cycle through the system? Which relationships need strengthening for the whole ecosystem to move?

In this framing, money is not merely fuel. It is pattern-making power. It can either reinforce silos or create the connective tissue that lets a system notice itself and respond.

That is why financing system change networks should not mean funding “networking” as a vague good. It should mean underwriting the specific capacities that make networks effective:

Sensemaking: the ability to define the problem collectively and update that definition as reality changes.
Brokerage: the ability to connect actors who would not otherwise coordinate.
Trust production: the slow, underappreciated work of making collaboration socially possible.
Rapid learning: the ability to test, observe, and adapt without requiring perfect plans.
Distributed ownership: the ability to prevent one actor from becoming the bottleneck or sole author of change.

These are not soft activities. They are the operating system of complex change.

The realist lens gives networks their missing discipline

One danger of network language is that it can become vague and romantic. People talk about ecosystem change, collective impact, or systems thinking, but without a disciplined way to learn, these efforts drift into pleasant meetings and inspiring diagrams. This is where realist synthesis offers a corrective. It gives network work a sharper question set: what works, for whom, in what circumstances, through which mechanisms, and with what side effects?

That question set is powerful because it avoids two common traps. The first trap is naïve universalism, the idea that one intervention should work everywhere if it is good enough. The second trap is pure relativism, the idea that because context matters, nothing general can be learned. Realist thinking says both are wrong. There are patterns, but they are conditional. Mechanisms are real, but they need contexts that let them operate.

This is especially important for funders who want to support systems change without becoming either micromanagers or passive donors. They need a theory of change that is humble enough to accept local variation and strong enough to guide action. That means designing portfolios around hypotheses, not just activities. It means funding groups not only for outputs, but for the quality of their learning loops.

A useful analogy is gardening rather than engineering. A gardener cannot force a seed to become a tree by applying more pressure. But neither does the gardener simply hope. The gardener prepares soil, adjusts water, responds to weather, removes weeds, and learns from each patch of land. The job is not control. The job is cultivating conditions.

Systems change funding works the same way. The question is not whether a network can guarantee an outcome. The question is whether it can create the conditions under which better outcomes become more likely, more repeatable, and more scalable.

The point of evaluation is not to punish uncertainty. It is to improve the system’s ability to learn from it.

What this means in practice: redesigning power, evidence, and time

If this thesis is right, then the real transformation is not only methodological. It is institutional. We need to redesign how power, evidence, and time are treated.

First, power. Networks do not automatically democratize anything. In fact, they can hide power under the language of collaboration. A truly useful funding model asks who sets the agenda, who gets heard, and who bears the cost of coordination. If the same dominant institutions always define the problem, the network is just a prettier version of the old hierarchy. Systems change requires intentional redistribution of voice, not just increased contact.

Second, evidence. The best evidence in complex settings is often mixed, local, and iterative. It includes quantitative signals, narrative feedback, implementation stories, and unexpected failures. A rise in one metric may matter less than a change in how actors interpret the problem. For example, if clinicians, community organizers, and policymakers start using the same language about barriers to care, that is an early sign of system alignment even before outcome metrics move.

Third, time. Networks need time not because they are inefficient, but because trust, shared language, and feedback loops take time to form. Short grant cycles often create a race to show progress before the work is mature. That pressures people to choose visible outputs over structural shifts. If the goal is durable change, funding must include patience as a design feature, not as a charitable exception.

A practical way to apply this is to ask, for any systems change effort:

What relationships must exist for the system to behave differently?
What would count as early evidence that the network is learning?
Which actors are missing from the table, and why?
What feedback loops currently keep the problem in place?
Which forms of uncertainty are essential, not a sign of failure?

These questions often reveal that the most important work is not the intervention you can point to, but the coordination capacity you are quietly building around it.

Key Takeaways

Stop funding only finished solutions. In complex systems, the earliest and most important asset is often the ability to coordinate learning across actors.
Treat relationships as infrastructure. Trust, brokerage, and shared sensemaking are not soft extras. They are the machinery of system change.
Evaluate mechanisms, not just outputs. Ask what changed, for whom, and under what conditions, rather than assuming one model should work everywhere.
Design for adaptation. Build funding arrangements that let people test, adjust, and share insights quickly instead of locking them into rigid plans.
Watch for hidden power. Networks can reproduce hierarchy unless agenda setting, voice, and resource allocation are intentionally distributed.

The real unit of change is not the intervention, it is the field around it

The most important insight from combining these two ways of thinking is that system change is not primarily about perfecting a tool. It is about changing the field in which tools operate. A well designed intervention in a broken ecosystem often underperforms. A modest intervention inside a learning-rich network can have outsized effects because it is amplified by trust, feedback, and coordination.

That is the deepest reason funding and realist evaluation belong together. One tells us how to resource the social architecture of change. The other tells us how to learn within that architecture without flattening complexity into false certainty. Together they point to a more honest practice: support the network, track the mechanisms, and let the system teach you what it needs next.

When we stop asking, “What single program should solve this?” and start asking, “What kind of collective intelligence does this problem require?” we open the door to a different kind of strategy. Not one that pretends complexity does not exist, but one that treats complexity as the terrain where meaningful change actually happens.

That is the reframing worth keeping: the goal is not to eliminate uncertainty before acting. The goal is to build the capacity to act intelligently inside uncertainty.