Why Organizations Cannot Optimize

Core Concepts

1. Bounded Rationality: The Architecture's Founding Assumption

Most mental models of organizations secretly assume a rational optimizer at the center: some executive (or some process) that, given enough time and data, can find the best solution. Herbert Simon's work demolished this assumption, and the implications have not stopped reverberating since.

Simon's bounded rationality framework, established in Administrative Behavior (1947) and extended through decades of subsequent research, makes a simple but consequential observation: decision-makers have limited cognitive capacity, incomplete information, and finite time. They cannot evaluate all alternatives, estimate expected utility for each, and select the optimal one. The machinery required for classical rationality does not exist in human minds, and it does not exist in organizations made of human minds.

The gap between normative decision theory and actual organizational behavior is not a failure of individual rationality. It is a predictable consequence of cognitive and informational architecture.

This is not a pessimistic finding. It is the theoretical foundation that makes organization theory possible: if individuals could costlessly optimize, coordination through hierarchy, specialization, and procedures would add no value. Organizations exist precisely because they do not.

2. Satisficing: Stopping When Good Enough

The behavioral response to bounded rationality is satisficing — a portmanteau of satisfying and sufficing that Simon introduced to name what decision-makers actually do.

Rather than exhaustively evaluating all options to maximize utility, decision-makers set a threshold of acceptability and stop searching once that threshold is met. The cognitive savings are not marginal. Maximization requires comparative evaluation of all known alternatives, utility estimation for each, and selection against an optimization criterion. Satisficing requires only: does this option clear the bar?

This efficiency is substantial. Satisficing reduces cognitive load and decision-making time compared to maximization strategies, and it does so in a way that is commensurate with the complexity and stakes of the decision. The threshold itself can be raised or lowered; aspiration levels adjust to circumstances.

The critical structural implication: if every decision-maker in an organization satisfices, the organization's behavior is a function of who is searching, what thresholds they hold, and what information reaches them — not a function of any global optimization process. Structure determines search, and search determines outcomes.

3. Information Cost as the Foundational Logic

Why does bounded rationality lead to the organizational forms we observe? The clearest answer comes from treating information processing as having a cost. When acquiring and processing information is expensive — in time, in cognitive effort, in coordination overhead — bounded rational behavior (satisficing, heuristics, delegation) emerges not as a failure but as a rational response to that cost structure.

Organizations respond to information costs in systematic ways:

• Delegation reduces how much information must travel to the top.
• Standard operating procedures avoid re-processing recurring decisions.
• Specialization reduces the information each unit must hold.

These are not arbitrary design choices; they are responses to the fundamental economics of information processing under cognitive constraints.

4. Nearly Decomposable Systems: The Architecture of Manageable Complexity

Simon's most architecturally consequential concept is nearly decomposable systems. A nearly decomposable system is one where:

• Interactions within subsystems are strong and frequent.
• Interactions between subsystems are weak and infrequent.

This structure is not accidental in complex organizations. It is the cognitive solution to a real problem: if every part of an organization needed to coordinate continuously with every other part, the information-processing requirement would exceed what any hierarchy of bounded-rational agents can handle. Near-decomposability makes that problem tractable by breaking it into nearly independent subproblems.

Fig 1

Each subsystem can operate as a relatively autonomous bundle of routines. Its members need to hold and process only local information. Cross-subsystem coordination happens at the boundary — infrequently, through well-defined interfaces. This is precisely the property that makes a microservice boundary sensible, a team topology rational, or a platform contract valuable. The decomposability of the system and the decomposability of the organization that builds it are the same problem in two registers.

5. Hierarchy as an Information-Processing Mechanism

Hierarchical organizational structures are not primarily about control or accountability. They are information-processing mechanisms that respond to bounded rationality by decomposing complex collective decisions through nested structures.

Each level of the hierarchy handles a subset of the organization's information. Local problems are resolved locally, without generating upward traffic. Problems that cross subsystem boundaries — that cannot be resolved with locally available information — escalate. The hierarchy is, at its core, an escalation filter.

Cyert and March's Behavioral Theory of the Firm (1963) extended this into a full organizational model. Firms are networks of information-processing agents where bounded rationality shapes structure through standard operating procedures, problemistic search (searching for solutions only when a problem is felt), and coalition formation. Organizational objectives emerge through negotiation, not optimization.

6. Delegation as an Information-Matching Problem

Delegation is typically framed as a control problem: how much authority to grant, how to monitor compliance, how to align incentives. The bounded-rationality framing reorients this completely.

Delegation depth and structure are shaped by where relevant information concentrates and where applicable expertise resides. Decisions requiring intensive information processing or specialized expertise are delegated to lower levels — not because those levels are trusted (though they may be), but because the information needed to make the decision is there, not above. Escalating such decisions to senior management would require compressing and transmitting information upward, losing fidelity in transit, and consuming scarce executive attention on problems where local actors already hold the relevant signal.

The converse also holds: decisions with high organizational impact but lower information intensity are retained at higher levels, where cross-subsystem visibility makes sense.

The effectiveness of this information matching determines whether a hierarchy economizes on cognitive load or creates bottlenecks. When decisions are assigned to levels that lack the information needed to make them well, the result is escalation chains, endless clarification requests, and decision latency — all symptoms of a structural mismatch, not individual incompetence.

7. Attention as a Scarce Organizational Resource

Bounded rationality does not merely limit individual decisions; it makes attention a scarce resource at the organizational level. Outcomes of organizational decisions are highly sensitive to the issues to which actors happen to be paying attention at any given moment.

This has structural consequences. Hierarchical structures and delegation mechanisms function partly as attention-allocation mechanisms. They determine whose attention goes where, routing the scarcest cognitive resource — executive attention — toward the organization's most critical functions. The organization's structure is, among other things, an attention budget.

This framing reframes several common organizational puzzles:

• Why do executives miss important signals? Because structure routes other signals to them.
• Why does a team make local decisions that seem obvious to escalate? Because the cost of escalation — consuming scarce attention above — is a real structural friction.
• Why do some problems grow until they become crises? Because no structural pathway directed attention to them early.

Standard operating procedures are a direct response to attention scarcity. By externalizing and codifying recurring decisions, SOPs remove those decisions from the pool competing for live attention. Routines encode organizational memory: patterned sequences of learned behavior that reduce cognitive load by automating responses to recurring situations, bypassing working memory constraints to enable coordination at scale.

8. Ecological Rationality: Heuristics as Adaptive, Not Deficient

The final concept reframes what bounded rationality means normatively. The classical interpretation treats satisficing and heuristics as second-best approximations: the organization does what it can given its limits, but would do better if those limits were relaxed.

Ecological rationality, developed by Gigerenzer and colleagues extending Simon's foundations, offers a stronger claim. Cognitive processes and simple heuristics are rational not in themselves, but in their degree of fit to environmental structures and task demands. What appear as biases in clean laboratory experiments may be superior strategies in real organizational environments, where information is structured, distributions have predictable shapes, and time pressure is constant.

Fast-and-frugal heuristics perform best in environments characterized by uncertainty, ill-defined problems, multiple weak information cues of unequal value, and many possible courses of action — the typical conditions of organizational decision-making. These are precisely the conditions where complex optimization approaches become impractical. The ecological rationality framework is normative in a specific sense: it specifies the conditions under which a heuristic is smart, rather than treating all heuristics as failures to optimize.

The practical consequence: heuristics and SOPs embedded in organizational structure are not workarounds for insufficient optimization capability — they are the rational response to the environment. Designing them out in favor of comprehensive analysis frameworks can actually reduce decision quality under typical organizational conditions.

---

Worked Example

Reading a Team Topology Decision through Bounded Rationality

Consider an engineering organization that runs a large e-commerce platform. It has accumulated four cross-functional teams, each owning end-to-end responsibility for a vertical slice of the product (catalog, checkout, fulfillment, recommendations). There is also a platform team. The CTO is considering whether to consolidate catalog and recommendations into a single team.

Here is how bounded rationality applies at each decision point:

Step 1: What information does each level actually hold?

The catalog and recommendations teams share some data pipelines and a search index. However, the coupling is weak and infrequent: recommendations consumes catalog data once a day in batch; the interface is a well-defined schema. This is a nearly decomposable boundary. The teams can operate as autonomous bundles of routines; cross-team coordination is the exception.

The CTO does not hold this coupling detail. It exists in the heads of the two tech leads and in the system's dependency graph. Any decision made without this local information is being made with compressed, lossy signal.

Step 2: Is this a decision to retain or delegate?

Per the delegation-as-information-matching principle: the decision requires intensive knowledge of data pipeline coupling, team cognitive load, and cross-team coordination frequency — all information that concentrates locally. This argues for keeping the decision with the tech leads, not the CTO. The CTO's role is to set the threshold (what organizational shape is acceptable, what team size is too large), not to resolve the coupling analysis.

Step 3: What satisficing threshold applies?

The CTO sets: "We should not have teams larger than twelve people, and we should not have inter-team dependencies that require more than two synchronization meetings per week." The tech leads apply that threshold locally. If the current structure clears it, no consolidation is needed. This is satisficing, not optimization: there is no attempt to find the ideal team topology across all possible configurations.

Step 4: What happens to attention?

The platform team's escalation queue is already full. Routing the consolidation decision upward would consume executive attention on a problem where locally-held information is better. By absorbing the decision locally within a satisficing threshold, the teams protect the CTO's attention for decisions where cross-subsystem visibility genuinely matters: roadmap conflicts, resource allocation across teams, or a new vertical that crosses all existing boundaries.

Outcome: The tech leads run the coupling analysis, find that recommendations can be served by an internal API contract rather than shared pipeline ownership, and propose a well-defined interface. No consolidation needed. The threshold is met. The decision is absorbed locally, attention is preserved upward, and the nearly decomposable boundary is strengthened rather than removed.

This is bounded rationality working well. The alternative — centralizing the decision and optimizing across all possible team configurations — would consume more cognitive resources and produce no better outcome.

---

Common Misconceptions

"Bounded rationality means organizations make bad decisions." This misreads the theory. Bounded rationality is descriptive, not pejorative. Simon's claim is that organizations cannot optimize — not that they fail systematically. Satisficing produces decisions commensurate with the cognitive effort invested. Under typical organizational conditions (uncertainty, time pressure, structured information environments), satisficing strategies often match or exceed the quality of attempted optimization, at far lower cognitive cost.

"Hierarchy is about control and accountability." Hierarchy is primarily an information-processing architecture. It distributes authority vertically to localize information processing and allow subsystems to operate as relatively autonomous routines. Control and accountability are properties that can be layered on top, but they are not the explanatory root. When a hierarchy malfunctions — bottlenecks, missed escalations, slow decisions — the fault is almost always informational, not motivational.

"Delegation is about trust." Delegation depth and structure are shaped by where information concentrates and where expertise resides, not by trust levels. Trust matters for how much monitoring you apply after the fact; it does not explain why decisions are placed where they are in the first place. Framing delegation as trust leads to the wrong diagnostics: when a delegated decision goes poorly, the instinct is to reduce trust (centralize), when the actual problem may be that the decision was placed at a level without the relevant information.

"Heuristics and SOPs are shortcuts that should be replaced by better analysis." Fast-and-frugal heuristics work because they exploit the structure of real organizational environments, not despite ignoring it. Replacing them with comprehensive optimization frameworks in environments characterized by uncertainty, many weak cues, and time pressure can reduce decision quality. The question is not "heuristic or analysis?" but "does this heuristic fit the environment it operates in?"

"Organizational structure can be designed from first principles." Formal organizational structures emerge from and constrain informal interaction patterns. Past interactions generate formal organizations, which in turn restrict future interaction possibilities. This means structure is historically contingent, path-dependent, and cannot be evaluated statically. A hierarchy that was efficient when it was formalized may be mismatched to the informal interaction patterns that have since evolved around it.