Research glossary

This glossary is for readers who want to understand how the pieces of my research fit together. It includes terms I have introduced or developed, such as unbounding rationality and strategic representations, as well as established ideas that do substantial work across my papers, such as bounded rationality and exploration and exploitation.

The definitions explain how I use each term. They are not claims that every term originated with me. The In my work lines point to the papers where the concept is developed or used most directly. Related entries mark distinctions that matter; they are not merely lists of synonyms.

Jump to: A · B · C · D · E · F · G · H · I · L · M · N · O · P · R · S · T · U · V · W

A

Adaptation

Adaptation is a change in a firm’s choices in response to experience or a changed environment. In my NK work, adaptation occurs as firms search across interdependent choices and move through a performance landscape. Whether they improve depends on the landscape, what they know, how they represent the problem, and how they search.

In my work: How much to copy?; Mental representation and the discovery of new strategies; How NK landscapes work; Government as landscape designer; Cognitive and structural antecedents of innovation; Search heuristics under multiple objectives

Aggregation

Aggregation is the process of combining information, predictions, representations, or preferences held by multiple people into an organizational judgment or choice. It is one of the three basic cognitive functions in my research, alongside search and representation. The result depends on the rule used: averaging, voting, delegation, unanimity, and polyarchy do not produce interchangeable decisions.

In my work: Organizational decision making; Strategic foresight; Organizations as artificial intelligences; Artificial intelligence and strategic decision-making; Distributed representations; AI and strategic decision-making; Revisiting the unitary actor assumption

Aggregating predictions and aggregating representations

Aggregating predictions combines people’s final estimates—for example, by averaging their forecasts. Aggregating representations combines the cues, categories, and causal beliefs that produced those estimates. The first is cheap and often effective because little information must be exchanged. The second requires deliberation but can change what group members notice and how they understand the problem.

In my work: Individual and organizational antecedents of strategic foresight; The power and limits of distributed representations

AI-augmented strategic decision-making

AI-augmented strategic decision-making uses AI inside a decision process rather than treating it as an oracle that supplies a finished answer. AI may generate alternatives, evaluate them, build richer representations, simulate competing views, or aggregate judgments. The relevant unit of analysis is therefore the designed human–AI system, not the standalone model.

In my work: Artificial intelligence and strategic decision-making; AI is revolutionizing strategic decision-making; Can AI do strategy? A dialogue and debate

Averaging

Averaging is an aggregation rule that takes the mean of members’ estimates or choice probabilities. It can cancel independent errors and is robust across many environments, but it works poorly when expertise is sharply uneven and the organization could identify the relevant expert. Averaging predictions is not the same as deliberating until members share a representation.

In my work: Organizational decision making; Strategic foresight; Distributed representations

B

Bounded rationality

Bounded rationality is Herbert Simon’s premise that decision-makers cannot know every alternative, consequence, or probability and cannot compute an unconstrained optimum. They act through incomplete representations and limited search. In my work, bounded rationality is not a vague claim that people make mistakes; it identifies specific limits that organization design, external representations, formal procedures, and AI may offset.

In my work: Strategic decision making; Mental representation and the discovery of new strategies; Strategic representations; Representational complexity; Organizations as artificial intelligences; External representations in strategic decision making; Unbounding rationality; Distributed representations

Breadth and depth of a representation

Breadth is the number of relevant dimensions or cues a representation considers. Depth is the detail with which it analyzes each one. A representation can therefore be broad but shallow, or narrow but deep. In the foresight studies, broader representations predict better judgments; taking a strategy course increases both depth and attention to industry and competitive concerns.

In my work: Strategic foresight; Learning strategic representations

C

Carnegie tradition

The Carnegie tradition sees organizations as information-processing devices composed of boundedly rational people. Organizations deal with the challenges they face through cognitive processes such as attention, search, representation, aggregation, learning, and cognitive division of labor. Heuristics, routines, aspiration levels, and satisficing make those processes workable despite limited knowledge and computation. My work makes representations and aggregation structures explicit and asks which cognitive limits AI may relax.

In my work: Organizational structure and mutual-fund performance; Organizational decision making; Mental representation and the discovery of new strategies; Strategic representations; Representational complexity; Organizations as artificial intelligences; External representations in strategic decision making; Unbounding rationality; Distributed representations

Related: Bounded rationality; Heuristic; Search

Causal Ladder and Delegation Ladder

The Causal Ladder classifies strategic tasks by the reasoning they require: analytic–predictive, intervention-oriented, counterfactual–extrapolative, and model-generative. The Delegation Ladder classifies the authority granted to AI: informational support, tactical recommendation, bounded operation, and strategic executive authority. The ladders answer different questions. The first asks what an AI can do; the second asks what an organization will let it do. Delegation depends on verified performance, bounded downside, and clear accountability, so an AI may receive substantial authority on a measurable task without reaching the highest level of causal reasoning.

In my work: Can AI do strategy?

Closed-form model

A closed-form model derives results analytically from equations, producing propositions or solutions that can be expressed without running a simulation. It exposes the logic linking assumptions to results, but tractability often requires strong simplifications. Closed-form and computational models are both formal models; the difference is how their implications are derived.

In my work: How formal models contribute to organization theory

Cognitive and structural antecedents

Cognitive antecedents are features of what organizational members know, such as the breadth and relevance of their prior experience. Structural antecedents are features of how those members are arranged, such as hierarchy and span of control. Their effects are task-specific: a structure or experience profile that encourages adoption may not support implementation under a new regime.

In my work: Cognitive and structural antecedents of innovation

Commission error

A commission error occurs when an organization accepts a bad project, practice, or opportunity. Requiring more agreement before approval generally reduces commission errors but increases omission errors. The desired balance depends on the relative costs of the two mistakes, so no decision structure is best independent of the task.

In my work: Organizational structure and mutual-fund performance; An efficient frontier in organization design; When consensus hurts the company

Competitive positioning

Competitive positioning concerns where a firm’s offering sits on attributes customers value and how that position performs against rivals. My work separates the choices of business policies from the market position those choices create. A set of internally efficient policies need not yield a position that remains viable once consumer trade-offs and competitors are considered.

In my work: Positioning on a multi-attribute landscape

Computational model

A computational model states assumptions as an executable procedure and derives implications by running it. Simulation makes it possible to study nonlinear interactions, heterogeneous actors, adaptation, and outcomes for which no tractable analytical solution exists. Its precision does not make it realistic by default: the model still needs transparent assumptions, validation, and robustness checks.

In my work: How formal models contribute to organization theory; How NK landscapes work

Computational strategy

Computational strategy is the project of teaching computers how to do strategy. Donald Knuth wrote that “Science is knowledge which we understand so well that we can teach it to a computer.” Trying to teach strategy to a computer forces us to state strategic reasoning precisely, exposes gaps in our theories, and gives us systems whose performance we can test and improve. Such systems must generate and evaluate strategies through processes of search, representation, and aggregation.

In my work: Artificial intelligence and strategic decision-making; Unbounding rationality; Can AI do strategy?; Can AI do strategy? A dialogue and debate

Consensus threshold

The consensus threshold is the amount of agreement required before an organization approves an initiative. A low threshold favors action; a high threshold blocks proposals unless many evaluators support them. Raising the threshold tends to reduce commission errors and increase omission errors.

In my work: Organizational structure and mutual-fund performance; When consensus hurts the company

Context similarity

Context similarity is the extent to which the same practices produce similar performance for two firms because their external settings are alike. It differs from firm similarity, which concerns the practices the firms already use. Broad imitation becomes dangerous when context similarity is low: faithfully copying an internally coherent system does not help if that system’s payoffs change across settings.

In my work: How much to copy?

Crowd-based idea selection

Crowd-based idea selection asks a group to choose among proposals, rather than merely generate them. Adding voters does not always improve selection because larger crowds reach progressively less accurate recruits. A smaller, carefully recruited crowd can therefore match or beat a larger one. The distinction between generating options and selecting among them also matters for AI-assisted strategy.

In my work: Limits to the wisdom of the crowd in idea selection

D

Decision boundary

A decision boundary divides a problem space into regions that receive different decisions, such as approval and rejection. Its position and shape follow from the cues, weights, and decision rule used to evaluate projects. For an individual, the boundary expresses the person’s internal representation. For an organization, it expresses both members’ representations and the rule used to aggregate them. A more effective boundary admits more high-quality projects while excluding more low-quality ones.

In my work: The power and limits of distributed representations

Decision-making structure

A decision-making structure specifies who evaluates an opportunity, whose judgments enter the decision, and how those judgments are combined. Examples include a single decision-maker, delegation to an expert, majority voting, averaging, unanimity, and polyarchy. Structure changes choices even when members and information stay fixed; it is therefore a causal component of organizational performance.

In my work: Organizational structure and mutual-fund performance; Organizational decision making

Delegation

Delegation assigns a decision to the person or system judged most knowledgeable about the case. It works well when expertise is diverse, the relevant expert can be identified accurately, and that expertise fits the environment. When routing is unreliable, voting or averaging may be safer because they do not depend on selecting the right expert.

In my work: Organizational decision making

Design space

A design space is the set of organizational structures that can be compared under a common model. Mapping it reveals dominance relationships, trade-offs, and efficient frontiers. The point is not to find one universal optimum but to identify which designs are feasible and efficient for a given balance of errors.

In my work: An efficient frontier in organization design

Dislodging effect

The dislodging effect occurs when an outside change knocks a firm off its current local optimum and restarts search. Imitating even a small part of another firm or facing unstable policy can lower performance immediately yet open a path to a better peak. The effect explains why copying and policy instability can help through search rather than through faithful transfer or superior policy content.

In my work: How much to copy?; Government as landscape designer

Distributed representation

A distributed representation is a collective model assembled from partial representations held by several people and, sometimes, artifacts. No member needs to hold the whole model. Its quality depends jointly on what members know and on how the organization aggregates them. Distribution can pool complementary knowledge, but it can also combine incompatible or systematically biased views.

In my work: Strategic representations; The power and limits of distributed representations

Dual search

Dual search is the simultaneous search over actions and over the representation used to evaluate those actions. Policy search changes what the firm does while holding its mental model fixed; representation search changes the model through which possible policies are perceived. Long-run performance often requires balancing the two because better actions cannot be found through a badly misspecified representation.

In my work: Mental representation and the discovery of new strategies

E

Efficient frontier in organization design

An efficient frontier contains organizational designs for which one type of error cannot be reduced without increasing another. Designs below the frontier are dominated: another structure makes no more omission errors and no more commission errors, with at least one strict improvement. The frontier turns exploration and exploitation into an explicit design trade-off rather than two qualities an organization can maximize independently.

In my work: An efficient frontier in organization design

Environmental complexity

Environmental complexity is the degree to which the effects of a choice depend on other choices. In landscape models it is represented by interdependence among decision variables. Complexity makes incremental improvement harder, copying more context-dependent, and simple representations less reliable. Its effect is contingent on the firm’s knowledge and decision process rather than mechanically harmful.

In my work: Government as landscape designer; Representational complexity

Exploration and exploitation

Exploration admits more uncertain or unconventional initiatives, increasing the chance of finding valuable opportunities but also of accepting bad ones. Exploitation screens more strictly, protecting the organization from bad projects but risking the rejection of good ones. In my organization-design work, the pair is operationalized through commission and omission errors rather than treated as loose labels for novelty and efficiency.

In my work: Organizational structure and mutual-fund performance; An efficient frontier in organization design

External representation

An external representation is a physical or digital artifact used to store, display, or manipulate information about a problem. Maps, diagrams, models, dashboards, and strategy frameworks extend memory and make relationships easier to inspect. Their value depends on fit with the task, usability, and malleability; putting a weak model on a slide does not make it cognitively useful.

In my work: Strategic representations; External representations in strategic decision making

F

Firm similarity

Firm similarity is the extent to which an imitator and a model firm already use the same performance-relevant practices. It is distinct from context similarity: two firms can operate in comparable environments while having very different activity systems. Existing overlap changes both what remains to be copied and how copied practices interact with the imitator’s current choices.

In my work: How much to copy?

Formal model

A formal model is a precise account of relationships among variables stated through equations, logical rules, or executable procedures. Formalization forces assumptions and mechanisms into the open and allows implications to be derived consistently. It can sharpen theory and guide empirical tests, but precision is not evidence: whether the model explains the world remains an empirical question.

In my work: How formal models contribute to organization theory

Four Ts of AI and strategy

The Four Ts organize research on AI and strategy into theories, tools, teaching, and terrains. Theories revisit strategy’s foundations; tools address strategic action; teaching concerns how people learn to lead with AI; terrains cover the competitive, institutional, ethical, and regulatory settings in which AI operates. The framework maps an emerging field rather than claiming settled boundaries.

In my work: Handbook of Artificial Intelligence and Strategy

Framework

A framework is a suggested representation for a class of problems. It tells a user which variables and relationships to consider and may generate prescriptions once populated with case information. Frameworks save cognitive effort, but they also direct attention and omit information. Their usefulness therefore depends on fit with the problem and on the user’s representational capability.

In my work: Strategic representations; Learning strategic representations; External representations in strategic decision making

G

Government as landscape designer

Government acts as a landscape designer when regulations and incentives change the payoffs attached to firms’ possible choices. Policy therefore does more than select winners or correct prices; it reshapes the environment through which boundedly rational firms search. A policy can help even when the state cannot identify the optimal firm strategy, especially when it trains search or dislodges firms from poor local optima.

In my work: Government as landscape designer

H

Heuristic

A heuristic is a simplified rule for choosing, evaluating, or searching when full optimization is infeasible. A heuristic can be effective because it focuses attention, reduces computation, or fits the structure of the environment. But its performance depends on what its simplification leaves out. My work treats heuristics as explicit procedures whose consequences can be compared, not as synonyms for bias or intuition.

In my work: Representational complexity; When less is more; Search heuristics under multiple objectives

Hierarchy and polyarchy

In the Sah–Stiglitz organization-design tradition, Hierarchy accepts a proposal only if every evaluator approves it, while Polyarchy accepts it if any evaluator approves it. The first rule suppresses bad projects but misses more good ones; the second reverses that error profile. I prefer Unanimity to Hierarchy for the all-must-approve rule because hierarchy has several other meanings in organization theory, including reporting levels and vertical authority.

In my work: Organizational structure and mutual-fund performance; An efficient frontier in organization design; When consensus hurts the company; Organizations as artificial intelligences; The power and limits of distributed representations; Revisiting the unitary actor assumption

Human–AI hybrid

A human–AI hybrid combines human and machine judgments or assigns them different parts of a task. Hybrid performance is not automatically better than either component: an added human can dilute a stronger model, and an added model can mislead a human outside its competence. The design problem is to allocate work according to comparative advantage, feedback quality, and accountability.

In my work: The strategic foresight of LLMs; Can AI do strategy? A dialogue and debate

I

Imitation breadth

Imitation breadth is how many of another firm’s practices an imitator copies. Copying more can preserve complementarities among practices, but it also imports choices that may not fit the imitator’s context. The effective breadth depends on interdependence, context and firm similarity, and time horizon. Imitation can improve performance by reproducing a superior system or simply by restarting search.

In my work: How much to copy?

Imitation radius

The imitation radius is the zone of better-performing competitors whose practices a firm can feasibly copy. A lagging firm benefits from imitation when an attainable model lies inside this radius; copying a distant frontier firm may require too many coordinated changes. The concept joins competitive position to the practical question of whether catch-up is possible.

In my work: When to innovate and when to imitate

Industrial policy as search intervention

Industrial policy is a search intervention when it changes how firms discover strategies, not only the payoff to a known choice. Stable incentives may pull firms toward a target; changing policies may train them to move or dislodge them from a local optimum. Whether this helps depends on environmental complexity and firms’ own search capabilities.

In my work: Government as landscape designer

Information aggregation

Information aggregation combines dispersed judgments about an opportunity into a decision. Its central question is not whether the organization has knowledgeable members but how their knowledge enters the final choice. Delegation exploits identifiable expertise; voting counts preferences or judgments; averaging pools estimates. Their relative performance changes with expertise diversity, routing accuracy, and environmental change.

In my work: Organizational decision making

Informedness

Informedness is knowledge of which variables are relevant before constructing a simplified representation. Many arguments for simple heuristics quietly assume that decision-makers can discard the right variables. When that assumption fails, simplicity may remove signal rather than noise, and a more complex representation can perform better even in a relatively simple environment.

In my work: A contingency theory of representational complexity

Innovation–imitation choice

The innovation–imitation choice asks whether a firm should search beyond the industry’s current frontier or learn from a better-performing rival. The answer depends on industry maturity and the firm’s competitive position. Frontier firms in young spaces have more reason to innovate; lagging firms in mapped spaces often have more to gain from feasible imitation.

In my work: When to innovate and when to imitate

Related: Imitation radius; Local search

Interdependence

Interdependence exists when the payoff from one choice depends on how other choices are configured. It creates complementarities and trade-offs: changing one practice can alter the value of several others. High interdependence makes landscapes rugged, broad imitation more valuable when contexts match, and competitive positions harder to infer from individual policies.

In my work: How much to copy?; Positioning on a multi-attribute landscape; Mental representation and the discovery of new strategies; How NK landscapes work; Government as landscape designer; Search heuristics under multiple objectives

Internal and mental representation

An internal representation is a representation held within a given decision-maker. If the decision-maker is human, the internal representation is a mental representation. Internal representation is the more general term because it applies to both humans and AIs. It determines which cues the decision-maker uses, how those cues relate to outcomes, and which strategies can be considered or evaluated.

In my work: Mental representation and the discovery of new strategies; Strategic foresight; Strategic representations; Representational complexity; Learning strategic representations; External representations in strategic decision making; Distributed representations

L

Learning strategic representations

Learning strategic representations means changing how a person organizes and weighs information about a strategic problem, not merely recalling more facts or frameworks. In the study of MBA education, students’ representations became deeper and paid more attention to industry and competition after the core strategy course. Those changes provide observable learning outcomes for strategy education.

In my work: Learning strategic representations

Less-is-more effect

The less-is-more effect occurs when a decision rule that deliberately ignores a group indicator predicts individual outcomes more accurately than a rule that uses it. The result can arise when the estimated relationship between group membership and the outcome is distorted by omitted variables or model misspecification. More information then worsens the prediction instead of refining it.

In my work: When less is more

Local optimum

A local optimum is a configuration that is better than every configuration reachable by the permitted small move, even though a better configuration may exist farther away. Local optima are what make search path-dependent: once incremental improvement stops, the searcher needs a larger jump, imitation, policy change, or a new representation to move again.

Local search

Local search evaluates nearby alternatives and moves when a neighbor improves performance. It is cognitively cheap and often realistic, but the path taken depends on the starting point and it stops at local optima. Much of my search research asks how imitation, organizational structure, multiple goals, government policy, representations, or AI change that path.

M

Majority voting

Majority voting accepts the alternative supported by more than half of the voters. It can be robust when expertise is broadly distributed and the organization cannot reliably identify the best expert. But larger voting crowds are not always better: marginal recruits may be less accurate, and correlated errors limit the gain from adding members.

In my work: Organizational decision making; Limits to the wisdom of the crowd

Malleability

Malleability is the extent to which users can change an external representation while working with it. A malleable diagram, model, or simulation lets users test alternative framings and update assumptions. That flexibility can aid search, but it also requires users who know what to change and how to interpret the result.

In my work: External representations in strategic decision making

Multi-attribute landscape

A multi-attribute landscape maps underlying business policies into positions on several product attributes and then maps those positions into market performance. It links operations, consumer trade-offs, and competition in one model. A policy change can improve operational efficiency yet move the product to a less viable competitive position.

In my work: Positioning on a multi-attribute landscape

Multi-objective decision-making

Multi-objective decision-making chooses among already known alternatives that have outcomes on more than one dimension. The problem is how to rank vectors such as profit and social performance when no alternative dominates on every goal. It differs from multi-objective search, where the alternatives themselves must first be discovered through path-dependent moves.

In my work: Search heuristics under multiple objectives

Multi-objective search

Multi-objective search discovers alternatives while performance has several dimensions. Preferences affect more than the final ranking: they must be implemented in a search heuristic that decides which move to try and when to stop. Two organizations with the same stated goals can therefore reach different parts of the performance frontier because they search differently.

In my work: Search heuristics under multiple objectives

Multi-objective search heuristics

Five rules show how the same social and financial goals can guide search differently. Maximize improves only financial performance. Combine improves their sum. Alternate pursues one goal until stuck, then switches. Penalize deducts shortfalls below a social threshold. Satisfice gives priority to financial performance only after the social threshold is met. Their trajectories, not just their preferences, produce different outcomes.

In my work: Search heuristics under multiple objectives

N

NK landscape

An NK landscape is a model of a problem with N choices and K interactions affecting each choice’s contribution to performance. When K is low, the landscape is smooth and incremental search works well; as K rises, the landscape becomes more rugged and contains more local optima. The model is a controlled way to study interdependence, adaptation, search, imitation, representation, and multiple objectives.

In my work: How much to copy?; Positioning on a multi-attribute landscape; Mental representation and the discovery of new strategies; A note on how NK landscapes work; Government as landscape designer; Search heuristics under multiple objectives

O

Oblique strategy

An oblique strategy is an alternative discovered by pursuing one objective that unexpectedly improves another. In the multiple-goal model, alternating attention between social and financial performance can escape a local financial optimum and find a strategy that matches or exceeds profit-only search while also improving social performance. The gain comes from the path of search, not from assuming the goals are naturally aligned.

In my work: Search heuristics under multiple objectives

Omission error

An omission error occurs when an organization rejects a good project, practice, or opportunity. Structures that require broader agreement tend to make more omission errors because any skeptical evaluator can block action. Omission errors operationalize the cost of excessive screening and connect organization design to exploration.

In my work: Organizational structure and mutual-fund performance; An efficient frontier in organization design; When consensus hurts the company

Organizational structure

Organizational structure is the arrangement of decision rights and reporting relationships through which members’ judgments become organizational action. In my work it is measured and modeled at the level of decision processes: who can propose, approve, reject, or aggregate a choice. Structure shapes the number of initiatives pursued and the types of errors made.

In my work: Organizational structure and mutual-fund performance; An efficient frontier in organization design; Cognitive and structural antecedents of innovation

Organizational utility

Organizational utility is a utility function recovered from the collective choices of members joined by a specified aggregation structure. It predicts organizational behavior without assuming that the organization literally has one mind. It is generally not the average of individual utilities: unanimity behaves approximately like their pointwise minimum, whereas polyarchy behaves approximately like their pointwise maximum.

In my work: Revisiting the unitary actor assumption

Organizations as artificial intelligences

Organizations can be understood as artificial intelligences because they are designed systems that search, represent problems, and aggregate information to pursue goals. The analogy is explanatory, not a claim that firms are computers. It reveals that many foundational ideas in organization theory—including heuristics, problemistic search, and distributed processing—were developed alongside or borrowed from AI.

In my work: Organizations as artificial intelligences

Related: Aggregation; Representation; Search

P

Performance benchmark

A performance benchmark is a standardized task and scoring rule used to compare strategic judgments, processes, or systems. Examples include out-of-sample forecasts, comparisons of human- and AI-generated plans, and tests of specific strategic capabilities. A useful benchmark supplies repeated feedback without collapsing the construct of good strategy into whatever is easiest to measure.

In my work: Unbounding rationality; Can AI do strategy? A dialogue and debate

Performance gradient

A performance gradient is a signal that tells a learning system whether a change improved its output and in which direction to adjust. Games provide clear gradients through wins and losses; strategy often provides delayed, noisy, and contested outcomes. Building credible gradients is therefore a prerequisite for AI systems to learn strategic work rather than merely imitate strategic language.

In my work: Can AI do strategy? A dialogue and debate

Policy, position, and performance

Policies are the underlying choices through which a firm configures its activities. Those choices jointly produce a market position on attributes customers perceive, and that position produces performance through demand and competition. Separating the two mappings prevents researchers from assuming that similar policies imply similar positions or that an operationally efficient position will remain competitively viable.

In my work: Positioning on a multi-attribute landscape

Problem space

A problem space is the set of configurations a decision-maker might consider for a problem. It is defined by the variables and possible values supplied by the representation, so changing the representation can change the space itself. Strategic search is difficult partly because the relevant actions and dimensions are not always known in advance.

In my work: External representations in strategic decision making; Strategic decision making

Problem-space satisfiability

Problem-space satisfiability is the share of configurations in a problem space that meet the decision-maker’s threshold of acceptance. Low satisfiability means acceptable solutions are rare, so search takes longer and the quality of the representation matters more. The construct links a problem’s structure to expected search effort.

In my work: External representations in strategic decision making

R

Random-utility model

A random-utility model represents the utility of an alternative as a systematic component plus a random term. The random term accounts for inconsistency in decision-making: the same person may not always make the same choice when facing the same alternatives because perception and cognition are imperfect. The model turns utilities into choice probabilities, which can then be combined through an organizational aggregation rule.

In my work: Revisiting the unitary actor assumption

Representation

A representation is a model used to generate predictions about a task environment. It selects variables, relationships, and possible actions while omitting others. Because decision-makers act through this model, representation is not a neutral description placed between analysis and choice; it determines what can be noticed, imagined, evaluated, and searched.

In my work: Mental representation and the discovery of new strategies; Strategic foresight; What makes a decision strategic?; Representational complexity; Organizations as artificial intelligences; Learning strategic representations; External representations in strategic decision making; Artificial intelligence and strategic decision-making; Unbounding rationality; Distributed representations; AI and strategic decision-making

Representational capability

Representational capability is a manager’s ability to choose an external representation that fits the problem, populate it with adequate information, and monitor whether it remains useful. A good framework cannot repair weak inputs or unnoticed mismatch by itself. This capability explains why the same visual or analytical tool can help one user and mislead another.

In my work: External representations in strategic decision making

Representational complexity

Representational complexity is the number and intricacy of variables and relationships included in a model of the environment. More complexity can reduce omitted structure but demands more data and risks learning noise. Its optimal level depends less on environmental complexity alone than on experience, uncertainty, informedness, and knowledge of the environment’s deep structure.

In my work: A contingency theory of representational complexity; Strategic representations

S

Satisficing

Satisficing searches until an alternative clears an aspiration or acceptance threshold rather than trying to identify a global optimum. The threshold determines when attention shifts or search stops. In the multi-objective model, the Satisfice heuristic pursues social performance until its threshold is met and then gives priority to financial improvement.

In my work: Search heuristics under multiple objectives; External representations in strategic decision making

Search

Search is the process of discovering possible actions when the alternatives and their consequences are not fully known. It is logically prior to choosing among a fixed menu: search determines which options ever reach evaluation. My work examines how landscapes, representations, organizational structures, imitation, policy, multiple goals, and AI alter the direction, breadth, and stopping points of search.

In my work: How much to copy?; Strategic decision making; Positioning on a multi-attribute landscape; Mental representation and the discovery of new strategies; How NK landscapes work; Government as landscape designer; Organizations as artificial intelligences; External representations in strategic decision making; Artificial intelligence and strategic decision-making; Unbounding rationality; Search heuristics under multiple objectives; AI and strategic decision-making

Search effort

Search effort is the number of configurations a decision-maker tries before stopping. It rises when acceptable solutions are rare, when the representation makes promising moves hard to identify, or when the acceptance threshold is demanding. External representations can lower effort by making useful information easier to extract, but a poor representation can send search in the wrong direction.

In my work: External representations in strategic decision making

Signal detection theory

Signal detection theory separates the ability to distinguish good from bad opportunities from the threshold used to act on that evidence. Moving the threshold changes omission and commission errors even when evaluative ability stays fixed. This makes it useful for organization design: structure can be analyzed as a system that shifts the organization’s effective acceptance threshold.

In my work: An efficient frontier in organization design

Social–financial frontier

Every strategy in the model produces two outcomes: social performance and financial performance. The social–financial frontier contains the strategies that are not dominated on these two dimensions. At a point on the frontier, improving one dimension requires reducing the other. A strategy inside the frontier is dominated because another available strategy performs at least as well on both dimensions and better on at least one.

In my work: Search heuristics under multiple objectives

Span of control

Span of control is the number of direct reports per supervisor. In an idealized hierarchy with a constant span, total employment reflects workers at every level, not only the bottom layer. The corrected measure therefore infers span from total employees and the number of supervisory levels rather than treating the lowest layer as the entire organization.

In my work: A note on calculating the average span of control

Related: Organizational structure

Specialist and generalist

A specialist learns deeply about one cue; a generalist learns a multi-cue representation of the whole problem. Specialists can excel when one cue dominates and can form a strong distributed representation when their partial models are aggregated well. A highly experienced generalist may outperform the group, but such broad individual experience is costly and uncommon.

In my work: Organizational decision making; The power and limits of distributed representations

Statistical discrimination

Statistical discrimination uses observed group membership to predict an individual’s unobserved outcome. Even if the group averages are estimated correctly, including the group indicator can reduce individual predictive accuracy when the underlying prediction model is misspecified. The result separates the use of more group information from the production of better individual forecasts.

In my work: When less is more

Related: Less-is-more effect

Strategic autonomy

Strategic autonomy is the discretion granted to a person or AI system to make or execute consequential strategic choices. It is an organizational property, not simply a technical capability. Autonomy expands when performance can be verified, losses can be bounded, objectives can be specified without easy gaming, and accountability can be assigned.

In my work: Can AI do strategy?

Strategic decision-making

Strategic decision-making studies the processes that produce consequential organizational choices. Strategic decisions differ from textbook choice problems because alternatives must often be discovered, probabilities are ambiguous, goals conflict, several people participate, and implementation matters. The field therefore studies search, judgment, politics, organizational structure, and aggregation rather than choice content alone.

In my work: Strategic decision making; What makes a decision strategic?; External representations in strategic decision making; Artificial intelligence and strategic decision-making; Unbounding rationality; AI and strategic decision-making

Strategic foresight

Strategic foresight is the ability to predict which present course of action will lead to a better future outcome. It is measured by predictions made before the outcome occurs, not by the persuasiveness of a retrospective story. Foresight depends on representations and aggregation; recent prospective evidence also shows that frontier LLMs can outperform experienced human evaluators on some uncertain venture predictions.

In my work: Individual and organizational antecedents of strategic foresight; The strategic foresight of LLMs

Strategic representation

A strategic representation is a model used to predict the consequences of consequential firm choices. It may be internal, external, or distributed. What makes it strategic is not its format but its role in high-stakes decisions. Its quality is judged by the decisions and predictions it supports, not by visual sophistication.

In my work: Strategic foresight; What makes a decision strategic?; Learning strategic representations; Unbounding rationality; Distributed representations

Strategist as architect

The strategist as architect designs the process through which strategic judgment is produced. As AI takes on more generation, evaluation, representation, and deliberation, the strategist’s work shifts toward choosing components, assigning roles, setting tests, supplying proprietary context, and deciding where human judgment remains necessary. The role is process design, not passive acceptance of model output.

In my work: AI is revolutionizing strategic decision-making

T

Task environment

The task environment is the actual problem faced by a decision-maker: the real relationship between possible actions, relevant conditions, and outcomes. A representation is an incomplete model of that environment. Keeping the two separate makes error visible and prevents a familiar framework from being mistaken for the world it describes.

In my work: Strategic representations; External representations in strategic decision making

Theoretical precision

Theoretical precision is clarity about assumptions, mechanisms, variables, and the implications that follow from them. Formal models increase precision because their transformations must be explicit enough for another person or a computer to reproduce. Precision helps expose contradictions and design empirical tests, but it does not make weak assumptions substantively correct.

In my work: How formal models contribute to organization theory

Related: Formal model; Verifiability

Third epoch of management education

The third epoch is the possible AI-driven reorganization of management education. The first epoch emphasized practitioner teaching; the second, following the mid-century reform of business schools, emphasized disciplinary research and analytical rigor. A third would arise if AI substitutes for parts of analysis and instruction while changing what schools teach, how they teach it, and what distinctive value they provide.

In my work: The effects of artificial intelligence on management education

Related: Four Ts of AI and strategy

Trade-off

A trade-off exists when improving one valued dimension requires sacrificing another. Some trade-offs are technological or competitive; others arise from a chosen representation, metric, or location on a frontier. Search matters because a firm inside the frontier can improve both dimensions, and a new technology or oblique path can move the frontier itself.

In my work: Positioning on a multi-attribute landscape; Search heuristics under multiple objectives

Training effect

The training effect arises in the model of government as landscape designer. A capable government can transfer knowledge to firms by periodically changing which policies it targets, giving firms repeated opportunities to learn about different parts of their strategy. The mechanism resembles rotating management trainees across functions. For incentives it can operate across different spans of intervention; for regulations it requires a narrow span because an inaccurate, wide-ranging regulation can lock firms into poor choices. This differs from the dislodging effect, which helps by restarting search even when government knowledge is limited.

In my work: Government as landscape designer

U

Unanimity

Unanimity accepts an alternative only when every participating evaluator approves it. It is the same structure that Sah and Stiglitz call Hierarchy. Unanimity protects against commission errors when good projects are rare or evaluators are inexperienced, but one negative judgment can block a valuable proposal. When applied to members’ choice probabilities, it makes organizational utility resemble the least favorable member’s utility.

Unbounding rationality

Unbounding rationality is the systematic relaxation of the cognitive limits that have long constrained strategic decision-making. AI can expand search, support representations too complex for unaided human cognition, and run aggregation processes at much larger scale. The term does not imply perfect rationality: data, objectives, verification, governance, and computation continue to impose bounds.

In my work: Unbounding rationality; Can AI do strategy? A dialogue and debate

Unitary actor assumption

The unitary actor assumption treats a firm as if it had one coherent preference ordering and chose accordingly. This is analytically convenient but hides disagreement among members and the structure used to resolve it. Organizational utility retains the tractability of a unitary representation while deriving it from individual preferences and an explicit aggregation rule.

In my work: Revisiting the unitary actor assumption

Usability of an external representation

The usability of an external representation is the extent to which helpful information can be extracted from it. A usable table, graph, map, or framework makes relevant patterns and relationships easy to detect. Usability is a characteristic of the representation, whereas representational capability is a characteristic of the person using it.

In my work: External representations in strategic decision making

V

Verifiability

Verifiability is the extent to which the quality of a strategic judgment can be scored credibly and repeatedly against an accepted standard. It requires more than eventual firm performance, which is delayed and confounded by implementation and luck. Forecasting tasks, simulations, comparative evaluations, and capability tests can provide nearer-term evidence while leaving room to audit what the metric misses.

In my work: Unbounding rationality; Can AI do strategy?; Can AI do strategy? A dialogue and debate

Verification infrastructure

Verification infrastructure is the set of benchmarks, simulations, audits, evaluation protocols, and governance arrangements that makes performance evidence credible before long-run outcomes arrive. It supports learning and justifies delegation. Poor infrastructure creates false confidence, especially when a metric is easy to game or captures only a narrow part of strategic quality.

In my work: Can AI do strategy?

W

Wisdom of the crowds

The wisdom of the crowds is the improvement that can result from combining multiple partly independent judgments. It depends on member accuracy, diversity, independence, and the aggregation rule. More people do not guarantee more wisdom: correlated errors, weak recruitment, or the addition of less accurate members can flatten or reverse the benefit.

In my work: Strategic foresight; Limits to the wisdom of the crowd; The strategic foresight of LLMs