When Plans Fail, Learning Begins: The Hidden Logic of Change in Complex Systems

Jul 24, 2026

9 min read

1 views

34%

The problem with believing in clean plans

Why do so many well funded programs begin with confidence, only to end in disappointment, confusion, or a report that says they “met activities but not outcomes”? The answer is rarely that the team did not work hard enough. More often, the deeper mistake is that they treated change like a straight line when reality behaves more like weather.

A plan assumes the world is stable enough for a causal chain to hold: if we do A, then B will happen, then C will follow. But in the real world, especially in public health, communities, organizations, and policy environments, the path from action to result is full of feedback loops, local conditions, and human interpretation. The same intervention can succeed in one place and fail in another, not because the idea changed, but because the context did.

This is where a more useful question emerges: not “Did the intervention work?” but “Under what conditions did it work, for whom, and why?” That shift sounds subtle, but it changes everything. It turns change from a verdict into an investigation.

The most important thing to understand about change is that it is not simply delivered. It is enacted, adapted, resisted, and amplified by context.

The hidden tension: causation versus context

Most institutions want certainty. Funders want clear metrics. Leaders want a model they can replicate. Policymakers want a program they can scale. These desires are understandable, but they create a dangerous bias toward oversimplification. When a social problem is complex, a simple causal story can be more comforting than accurate.

That is why theories of change, logic models, and evaluation frameworks are both powerful and limited. They help make assumptions visible. They clarify intended pathways. They force teams to ask how resources become activities, activities become outputs, and outputs become outcomes. Yet they often imply a neatness that reality refuses to provide.

A public health campaign, for example, may distribute information about vaccination. On paper, the causal chain looks straightforward: knowledge increases, hesitancy decreases, uptake rises. In practice, uptake may depend on trust in institutions, transportation access, clinic hours, social norms, language, prior experiences of discrimination, and whether local leaders endorse the effort. The information matters, but it is only one actor in a crowded field.

This is the deeper tension at the heart of change work: the need for generalizable guidance collides with the fact that outcomes are produced in specific, living contexts. If we insist on a universal recipe, we miss how reality actually behaves. If we give up on structure altogether, we drift into anecdote and improvisation. The challenge is to build a framework that is disciplined without being delusional.

Why realist thinking is not just another evaluation method

A more mature approach starts by treating programs as theories about the world, not just bundles of activities. Every intervention implicitly says something like this: “If we introduce this mechanism in this context, it will trigger these responses and produce these outcomes.” That sentence is richer than a simple cause and effect statement because it acknowledges mechanisms and context.

Mechanisms are not the program itself. They are the underlying responses it tries to activate. A reminder text does not reduce missed appointments because text messages are magical. It works only if recipients see the message, understand it, trust it, can act on it, and are not blocked by other constraints. The same text may help one patient and do nothing for another.

This is why a realistic synthesis of evidence is so valuable in public health and beyond. It asks what kinds of settings make a mechanism fire. It asks what blocks the chain. It asks what adaptations preserve the core function while changing the form. In other words, it moves evaluation away from “Did the average effect exist?” toward “What is the explanatory pattern across different conditions?”

That difference matters because many failures are not failures of the idea. They are failures of fit.

Consider a job training program for unemployed young adults. A conventional evaluation might ask whether participants found jobs. A realist lens asks a fuller set of questions: Did the program increase confidence, social support, or employer access? Did those mechanisms matter differently in neighborhoods with weak labor markets? Did transportation barriers or childcare constraints prevent participation? Did the program help only those who were already closest to employment?

Now we are no longer just measuring outcomes. We are mapping the causal ecology.

Good change work does not merely ask whether something caused an outcome. It asks how an outcome was produced through the interaction of design, context, and human response.

A useful mental model: from arrows to ecosystems

Traditional theories of change often look like a row of arrows. Input leads to activity, activity leads to output, output leads to outcome. This is useful as a starting point, but it can create the illusion that the world is a conveyor belt. Real systems are not conveyor belts. They are ecosystems.

In an ecosystem, a change in one species alters the behavior of others. A small shift in rainfall changes plant growth, which changes grazing patterns, which changes soil quality, which changes the whole system again. Human systems behave similarly. A policy change affects incentives. Incentives reshape behavior. Behavior changes norms. Norms feed back into policy acceptance. The loop never stops.

This suggests a better model for thinking about change: the Intervention Ecosystem Model.

It has four parts:

Intent: What are we trying to change?
Mechanism: What psychological, social, or institutional response should drive the change?
Context: What conditions help or block that response?
Feedback: How does the system react once the intervention begins?

This model is useful because it prevents a common mistake: assuming that an intervention is identical to its effect. A mental health program may be designed to increase help seeking, but the mechanism could be undermined by stigma, privacy concerns, or mistrust. A nutrition initiative may aim to shift food choices, but if the local food environment is unchanged, the mechanism remains weak. A policy can be elegant and still fail if the system pushes back.

The best analogy is not a machine but a conversation. When you speak, the listener interprets, resists, agrees, or redirects the exchange. Change works the same way. Interventions are messages to a system, and systems answer back.

The practical payoff: better design, not just better evaluation

This way of thinking is often discussed as an evaluation philosophy, but its biggest value may be design. If you understand change as context dependent, you stop asking, “How do we scale one perfect intervention?” and start asking, “How do we preserve the mechanism while adapting the form?”

That distinction is crucial. Scaling is not copying. A successful program is not a sacred object to be reproduced identically everywhere. It is a set of functions that must be translated into different environments. The core question is which elements are essential and which are simply one workable expression of the idea.

For example, if a youth mentoring program succeeds because it creates reliable adult relationships, then the key ingredient is not the particular workbook or weekly format. The key ingredient is trustworthy relational continuity. In one community, that might require school based mentors. In another, it might require after school groups led by local volunteers. The mechanism stays, the delivery changes.

This is where realist synthesis and theories of change become complementary. One offers a disciplined map of assumptions. The other asks how those assumptions behave in the real world. Together they help teams move from “implementation fidelity” to functional fidelity: preserving what makes the intervention work, even when the surface form changes.

That idea is especially important in public policy, where imported solutions often fail because they carry their original context inside them. A successful program from one city may depend on civic trust, funding patterns, staffing norms, and administrative capacity that do not exist elsewhere. Copying the surface without transferring the supporting conditions is like transplanting a tree without soil.

The most important insight: failure is data about the system

One of the most liberating shifts in this field is to stop treating failure as a simple indictment. In complex systems, failure often tells you more about hidden conditions than success does. If a program works only in some places, that is not a nuisance. It is a clue.

Maybe the intervention needs a stronger trust bridge. Maybe it depends on frontline staff having enough autonomy to adapt. Maybe the outcome is gated by a resource that was never named in the original plan. Maybe the theory was right, but the mechanism was too weak to overcome competing forces.

This is why high quality change work requires humility. Not the performative humility of saying “systems are complex,” but the operational humility of building in learning loops. Test assumptions early. Look for heterogeneity. Ask where the intervention is thriving, where it is stalling, and what the local conditions reveal. Treat variation as information, not noise.

A hospital trying to reduce readmissions might discover that reminder calls reduce returns for one subgroup but not another. Instead of concluding the program failed, it can ask whether language barriers, medication cost, caregiver support, or discharge comprehension differ across groups. Each answer is a design opportunity. The evaluation becomes a map of leverage points.

In complex systems, the question is not whether something works everywhere. It is what it takes for it to work somewhere, and how those conditions can be recognized and recreated.

Key Takeaways

Stop asking only whether an intervention works. Ask under what conditions it works, for whom, and through which mechanism.
Separate the mechanism from the delivery form. Preserve what causes change, but remain flexible about how it is implemented.
Treat context as causal, not decorative. Culture, trust, capacity, timing, and incentives do not just surround an intervention, they shape its effect.
Use failure diagnostically. When outcomes differ across settings, look for the hidden system conditions that explain the variation.
Build feedback into the plan. A good theory of change should be revisable, not just presentable.

From tidy theories to living systems

The deepest flaw in many change frameworks is not that they are wrong. It is that they are incomplete in a particularly seductive way. They are good at showing intention, but weaker at showing interaction. They explain how we hope the world will respond, but not how the world actually negotiates our efforts.

A richer approach begins with a different stance: interventions are hypotheses about mechanisms in context. That means they are always provisional. Their purpose is not merely to justify action, but to learn how action becomes effect. The theory of change is not the answer. It is the first draft of the question.

This reframes evaluation as a discipline of curiosity. Instead of asking whether a program met a target, we ask what the target concealed. Instead of treating standardization as the path to quality, we recognize that quality often depends on intelligent adaptation. Instead of pretending that context is an obstacle to control, we accept that context is the medium through which change happens at all.

If you remember only one thing, let it be this: change does not travel in a vacuum. It moves through relationships, institutions, beliefs, and constraints. The most effective systems are not those with the most elegant plans, but those that can see the world clearly enough to revise the plan without losing the purpose.

That is the real promise of combining causal thinking with realist thinking. It does not make the world simple. It makes us better at working with its complexity. And once you see change that way, you stop looking for perfect programs and start building smarter conversations with reality.