Abstract Research Summary Modular systems play a central role in technological innovation. Such systems emerge when interdependencies among modules in a complex system are isolated through interfaces. While early seminal work highlighted the importance of interface design, subsequent research on modularity has largely overlooked it. We develop a model that treats interfaces as a set of design choices, separate from module choices. This model elucidates the mechanisms through which interface design influences system performance and identifies novel strategies for sequencing the search of interface and module designers to improve outcomes. The framework has implications not only for standalone innovations but also for the design of standards in platforms and ecosystems. Ultimately, it demonstrates that interface design is as much a strategic challenge as it is a technical one. Managerial Summary Managers increasingly rely on modular designs to enable innovation, yet often overlook interfaces as a strategic lever. This study shows that actively designed and periodically updated interfaces can coordinate interdependencies without constraining decentralized search, enabling modular systems to approach the performance of integrated designs. Crucially, sequencing matters: allowing modules to evolve before introducing interfaces improves long‐run performance, as early experimentation generates knowledge that interfaces can later build upon. Finally, infrequent interface redesign is sufficient to sustain coordination, reducing the need for continuous adaptations. Overall, interfaces should be treated as evolving strategic choices that shape innovation trajectories in products, platforms, and ecosystems.

We investigate how managers’ assumptions about noise in performance feedback impact adaptive search in complex environments. Using computer simulation, we demonstrate that rapid learners—managers who underestimate feedback noise and, therefore, revise their beliefs more aggressively—make more commission errors early on but can quickly self-correct these errors through additional experimentation. Over time, rapid learners reduce commission errors by increasingly detecting feedback with a high signal-to-noise ratio, and this prompts resampling and belief refinement, allowing them to outperform even unbiased Bayesian learners in environments that are of greater complexity. In contrast, cautious learners—managers who overestimate feedback noise and, therefore, take longer to infer confidently the superiority of better alternatives—make fewer commission errors but more omission errors. Cautious learners perform poorly in complex environments but are effective in simpler search spaces, in which sensitivity to subtle performance differences is advantageous. Our findings provide boundary conditions for the learning literature’s recommendation of gradual belief updating under uncertainty. Specifically, we show that underestimating feedback noise—and, thus, learning rapidly—can strike an effective balance: it encourages broad exploration early on, enabling the refinement of beliefs as managers encounter and repeatedly sample exceptionally high-performing solutions. Supplemental Material: The online appendix is available at https://doi.org/10.1287/stsc.2024.0246 .

Research Summary Studies using archival organizational structure data are not as prevalent as one might expect for such a critical strategy topic. We seek to facilitate more studies in this domain by introducing a novel, hand-collected dataset of top management team compositions of S&P 500 firms between 1993 and 2020. Alongside providing the original role titles, we use generative Artificial Intelligence (AI) to categorize executives' titles into 6 role groups and 12 hierarchical levels, enabling easier comparisons of structures across and within firms. Our findings not only align with prior research but also offer insights into industry-specific structural changes, functional distributions within organizations, and the evolution of executive roles. This work also highlights the potential of generative AI as a tool to empirically investigate key strategy questions.Managerial Summary One of the most important decisions senior managers make pertains to defining their firms' organizational structures. However, obtaining data on firms' structures can be challenging due to difficulties in accessing data and comparing structures across firms. In this paper, we develop a novel dataset of top management team compositions of S&P 500 firms between 1993 and 2020. Alongside providing the original names and job titles, we use generative Artificial Intelligence (AI) to categorize executives' titles into 6 role groups and 12 hierarchical levels, allowing easier comparisons of structures across and within firms. We hope that this new dataset will spur greater scholarly interest in organizational structure, offering insights into how firms are structured and the implications of these structures.

This study extends the behavioral theory of the firm by examining how functional subgroups within organizations influence strategic decision-making differentially across hierarchical levels. We argue that a larger proportion of managers in a specific functional subgroup within an executive team causes this group to focus the organization's strategic agenda on those issues that this functional subgroup considers most important. However, within a functional subgroup, the issues to which managers pay attention can vary between hierarchical levels, with senior managers' attention shaped by higher-level strategic schema, and lower-level managers' attention shaped by local, more operational stimuli. We test our theory using data from 1,064 U.S. hospitals, 6,218 executives, and Medicare-adjusted pricing for hospitals' shoppable services. Our analysis focuses on two key functional subgroups: the revenue subgroup, which focuses on financial sustainability, and the medical mission subgroup, focused on quality and accessible healthcare. We find that a higher proportion of revenuefocused managers at the system level is associated with higher insured pricing. In comparison, a higher proportion of system-level medical mission managers is associated with lower pricing. Interestingly, at the hospital level, a greater proportion of medical mission managers correlates with higher insured pricing, likely reflecting efforts to secure additional resources. Posthoc analyses further clarify the mechanisms behind these relationships. Our results highlight that functional subgroups are not monolithic; their managers' perspectives can vary with their relative position in the organizational hierarchy, and this can provide an important mechanism of conflict resolution.

In this study, we propose LLM agents as a novel approach in behavioral strategy research, complementing simulations and laboratory experiments to advance our understanding of cognitive processes in decision-making. Specifically, we reproduce a human laboratory experiment in behavioral strategy using large language model (LLM) generated agents and investigate how LLM agents compare to observed human behavior. Our results show that LLM agents effectively reproduce search behavior and decision-making comparable to humans. Extending our experiment, we analyze LLM agents' simulated "thoughts," discovering that more forward-looking thoughts correlate with favoring exploitation over exploration to maximize wealth. We show how this new approach can be leveraged in behavioral strategy research and address limitations.

We explore the role of generative artificial intelligence (AI), specifically ChatGPT, in the classroom to support learning and complement assignment evaluations. Incorporated into a Spring 2023 course, we assessed the tool from students' and instructor's perspectives. Initial student hesitancy was overcome as the technology's integration enhanced learning experiences and equipped them with valuable skills for future careers, leading to strong student advocacy for AI inclusion in curricula. From the grading perspective, the AI's stricter evaluations and occasional overemphasis on keyword inclusion highlighted areas of improvement. Despite this, AI has the potential of supplementing human grading, boosting fairness and consistency. Our experience emphasizes AI's potential in education, the need for teaching these non-trivial skills, and important areas for consideration.

Modular systems are increasingly important in the context of technology product designs. A modular system emerges when interdependencies across modules are isolated by an interface. While Baldwin and Clark (2000) emphasized that interfaces need to be designed, the modularity literature in management has either assumed that interfaces simply stabilize interacting elements or waved away the need for an interface entirely by assuming that communication and coordination across modules is costless. We develop a novel extension of the NK model (i.e., endogenous NK model) and conceptualize interfaces as a set of design choices that are separate from the choices about the modules. We use the model to explore the performance implications of various implementation strategies for the deployment of an interface. We find that interface design choices play a critical role in the performance of the modular system and affect its advantage relative to an integrated system in the short and, under certain conditions, long run. Importantly, we introduce the notion of an active interface and show that learning at the module level prior to the introduction of an interface is essential for the successful design of an interface and that such learning complements prior architectural knowledge.

The organizational design literature stresses the importance of organizational structure to understand strategic change, performance, and innovation. However, prior studies diverge regarding the conceptualizations and operationalizations of structure. Organizational structure has been studied as an (1) arrangement of activities, (2) representation of decision-making, and (3) legal entities. In this point-of-view paper, the three prominent perspectives of organizational structure are discussed in terms of their commonalities, differences, and the need to study their relationship more thoroughly. Future research may not only wish to integrate these dimensions but also be more vocal about what type of organization structure is studied and why.

Product innovation can result from the novel design and combination of product components as well as from changing the underlying architecture: that is, the way components interact with each other. Even though previous studies have shown that architectural change can constitute a powerful source of innovation, little insight exists on how organizations should engage in architectural search itself. In this paper, using computer simulation, we explore underlying mechanisms of architectural search. We find that contrary to search for component combinations, architectural search provides greater performance improvements the narrower the search scope, regardless of product complexity. Moreover, our theory and findings suggest a more differentiated typology of architectural innovation. Although narrow architectural search often leads to pure architectural innovations that do not require substantial component changes, broader architectural search often leads to composite architectural innovation (i.e., architectural innovations that typically render existing component designs suboptimal but allow for new high-performing component combinations to arise). Lastly, although narrow architectural search outperforms broad architectural search in the long run, in the short run broad architectural search can have performance advantages.

How can entrepreneurs develop a high performing product in uncertain markets? The entrepreneurship literature suggests to continuously experiment with prototypes to learn from performance feedback. However, there is a lack of guidance about how feedback from experimentation should be used in order to find high performing product designs. To answer this question, we develop a formal simulation model and explore how entrepreneurs'

The mental models held by managers -- their cognitive representations of the real word -- have an important effect on decision making and influence the way experiences are encoded. Mental models contain performance expectations of different actions, and higher-level beliefs of contingencies, i.e., whether prior experiences in one setting are applicable in other settings, a contingency model. While a growing literature has explored how first-order learning of performance expectations is affected by fixed contingency models that do not reflect reality correctly, relatively little is known how second-order learning with respect to the higher-level contingency model affects decision outcomes. Using a computer simulation built on prior theory and models of learning and cognition, we find that a simple mental model that is refined over time tends to lead to higher performance than a complex mental model that is simplified over time. Our findings suggest that the ubiquity of relatively simple mental models may not only be a reflection of cognitive limitations, but also an evolutionary advantageous outcome of a dynamic learning process.

The need for venturing out to distant states is a well-established insight of the behavioral theory of the firm. Previous literature has paid particular attention to the need to complement local experimentation with exploratory search for solutions that involve a multitude of changes, especially when decision problems are complex. Less attention has received the notion that exploration is not a process that occurs instantaneously but rather involves a series of interim steps. An important distinction therefore is to be made between discovery of a distant opportunity and the local learning efforts of the underlying activities that are required. We argue that the discovery of a distant opportunity may serve as a goal that then may inform a firm’s local search. In a computer simulation, we explore how a decision maker’s consideration of a distant goal in local learning shapes a firm’s search path. In particular, we examine how a firm’s level of goal myopia and the goal setting scope affect firm performance. We find that goal myopia to some extent is necessary and helpful in navigating through complex decision problems. We also find that a bounded goal setting scope is preferable in complex environments as firms may in fact reach such goals and more frequently update their goals along the way. In contrast, when firms set very attractive but also very distant goals, they are more likely to get stuck along the way with mediocre strategies.

This chapter reviews recent advances in the NK modeling literature conceptualizing organizational change and innovation as a search over a complex landscape. It discusses both strengths and limitations of this perspective and delineates potential for future research directions. The key argument is that the NK model in its traditional form may be exhausting the theoretical insights that it can provide to the field. However, substantial modifications and extensions of the NK model or new classes of landscape models may provide fresh perspectives. Specifically, we consider the modeling efforts that endogenize the landscape construction as the next frontier in this literature. We also discuss several recent studies that incorporate various extensions of the NK model and allow for agent-driven changes to the landscape.

We examine an intriguing paradox regarding whether interdependencies in an organization's activity system enable or hinder strategic renewal the incremental process through which an organization continuously adapts to the environment and explores opportunities to invoke change in its activity choices and outputs. One research stream, the "inertial view," argues that the pervasiveness of interdependencies among activities increases inertia, which inhibits strategic renewal. Another research stream, the "adaptive view," argues that the pervasiveness of interdependencies among activities allows for a rich flow of resources and information, which enables strategic renewal. In this article we argue that both views provide important insights but arrive at conflicting conclusions because they focus on different dimensions of interdependency. To resolve this paradox, we distinguish between an activity system's interdependency patterns and interdependency rules. We propose that the dimensions of the interdependency pattern set the context in which the dimensions of the interdependency rules guide the exchange of resources and information among interdependent activities. Integrating these two components of an activity system's interdependency design leads to a "dual understanding" of interdependency as both pattern and rule and helps explain how the inertial forces of interdependency patterns may be overcome by putting appropriate interdependency rules in place.