Rethinking organizational decision-making: The emerging roles and tasks of generative artificial intelligence

You’re listening to “Rethinking organizational decision-making: The emerging roles and tasks of generative artificial intelligence,” by N. Schulte and D.K. Kanbach. Published in 2026.

Abstract.

The rapid diffusion of generative artificial intelligence (GenAI) has triggered a transformative shift in how organizations approach decision-making. Despite growing enthusiasm and widespread adoption across industries, GenAI’s specific tasks and roles, and the ways in which they shape the interplay of human cognition and algorithmic enhancement in organizational decision-making, remain insufficiently understood. Addressing this gap, this study conducts a systematic literature review that identifies 68 relevant publications to synthesize and advance current knowledge on the integration of GenAI into decision-making. The study identifies 53 tasks performed by generative applications, aggregates them into 18 task categories, and maps these tasks and categories onto six recursive decision-making components: attention, intelligence, design, choice, implementation, and feedback.

Building on the harmonization and translation of these tasks, we propose a typology comprising six active GenAI roles and one collaborative human-AI role. We then develop a processual framework that specifies how and when GenAI is embedded within organizational decision-making processes, delineating how generative applications support, augment, or co-perform decision-making activities. Our findings reveal a fragmented application landscape and highlight the limited integration of GenAI in the choice phase of organizational decision-making. By offering a structured typology and a processual conceptual framework, this study clarifies the evolving interplay between human decision-makers and generative technologies. In doing so, it provides a foundation for theory-advancing research and for more explicit and actionable managerial practice.

Communicated by Christopher Albert Sabel.

Extended author information available on the last page of the article 1 Introduction

With the release of ChatGPT in November 2022, generative artificial intelligence (GenAI) began to reach organizational frontlines, rapidly entering workplaces and reshaping how work is performed. With steadily rising numbers of GenAI users and the rapid adoption of generative applications across the workforce, organizations are eager to capitalize on these socio-technological developments. Already deployed across multiple industries and functional domains, GenAI exerts a substantial impact on organizational efficiency, creativity, and agility. As GenAI diffusion permeates organizations and transforms core processes, tasks previously insulated from automation are increasingly affected, prompting firms to reconsider how work distributions and decisions are made.

As the generative capabilities of GenAI trigger changes in human cognition, such as shifts in cognitive effort, creative diversity, and reasoning, they extend organizational decision-making beyond traditional decision-making logic. Whereas decision-making was previously mediated by predictive artificial intelligence (AI), whose algorithmic support was limited to enhancing information provision and enabling decision-makers to partially alleviate their cognitive constraints, GenAI's capabilities may now catalyze broader epistemic shifts within decision-making. In doing so, the integration of GenAI introduces novel tensions around task and role clarity between human decision-makers and generative algorithmic systems.

Given the imperative to reap the best of both human cognition and algorithmic decision-making, clarity regarding GenAI's roles, tasks, and influence within organizational decision-making is needed.

As organizations increasingly rely on machine intelligence to support human decision-makers, tasks such as solution generation and outcome forecasting are progressively delegated to generative models. In this context, GenAI applications are frequently conceptualized and deployed as cognitive assistants, task agents, or equivalent team members supporting human decision-makers. Yet the integration of generative tools into organizational decision-making remains nascent and may be accompanied by risks and limitations that challenge the authority of GenAI in decision contexts. By shaping behavioral tendencies of decision-makers, such as directing attention or favoring high-probability solution patterns, GenAI may exert a distinct influence on cognitive, procedural, and governance mechanisms within organizational decision-making.

As a result, questions of distinct task and role allocation in human-AI decision-making remain unresolved.

Against this backdrop, a more precise understanding of GenAI’s roles and tasks is required to evaluate the interplay between human decision-makers and generative, algorithmic applications. To capture the nuances and dynamics through which GenAI shapes organizational decision-making, this study is therefore guided by two research questions: “What roles and tasks are performed by GenAI in organizational decision-making?” and “How does GenAI alter cognitive, procedural, and governance mechanisms in organizational decision-making?”.

To answer these questions, this study conducts a systematic literature review (SLR) that aligns theoretical relevance with practical significance. Following the replicable sequence of the methodology, we identify and analyze 68 relevant articles that highlight the different tasks and roles performed by GenAI within organizational decision-making. Based on the reviewed literature, we recognize 53 tasks performed by GenAI, 18 corresponding task categories, and six distinct roles and one collaborative role of generative applications. From these insights, we develop a sharpened understanding of GenAI-mediated decision-making and conceptualize a GenAI task and role typology along six core components of organizational decision-making: attention, intelligence, design, choice, implementation, and feedback.

In addition, we propose a processual and recursive framework outlining how GenAI intervenes across these components, enabling a more nuanced assessment of the interplay of human cognition and algorithmic enhancement within decision-making. In doing so, we examine how GenAI reshapes decision-making processes and how decision-makers may benefit from, or be adversely affected by, the integration of generative tools.

Theoretically, we contribute to the discourse on algorithmic agency, hybrid human-AI cognition, and socio-materiality. In mapping GenAI’s tasks and roles along the decision-making process, we highlight that bounded rationality is alleviated only selectively when GenAI is incorporated into organizational decision-making and, depending on the phase of decision-making, generative applications excel at formal rather than substantive rationality. Furthermore, by reviewing the current body of literature, we identify future research directions that invite reevaluation of established theories (e.g., attention-based view, bounded rationality) in the era of generative decision support systems. Practically, we delineate concrete mechanisms for leveraging GenAI’s procedural capabilities in decision-making.

In doing so, we propose how GenAI can be integrated into organizational decision-making in a manner that draws on the strengths of both human cognition and algorithmic decision-making. These insights are intended to assist practitioners in evaluating how and when to integrate GenAI.

2 Theoretical background 2.1 Organizational decision-making

Five decades ago, Mintzberg et al. (1976) identified 25 distinct decision-making processes, which can broadly be grouped into three categories: individual decision-making, group decision-making, and organizational decision-making. Individual decision-making, rooted in cognitive psychology and often relying on the verbalization of an individual’s thought processes, and group decision-making, typically examined in laboratory settings within social psychology, are, however, of limited relevance to this study, as they do not explicitly theorize the structural and processual foundations of decision-making. Organizational decision-making, by contrast, has evolved from early analyses of unstructured corporate decision processes into a rich body of research conceptualizing how decisions are formed, shaped, and enacted.

In line with neoclassical economic theory, early decision-making research assumed that choices are made under conditions of perfect information. Scholars associated with the Carnegie Tradition, however, challenged this assumption by emphasizing the inherent incompleteness of information in decision contexts and introducing the concept of bounded rationality—cognitive limitations that lead decision-makers to seek satisfactory rather than optimal solutions. Thereby treating decisions as problem-solving under environmental constraints of uncertainty, complexity, and time limitations. Building on this foundation, Simon's (1947, 1960) seminal work, later extended by Eisenhardt and Zbaracki (1992), proposed an iterative and recursive three-phase model of managerial decision-making, comprising intelligence, design, and choice.

In the intelligence (or problem identification) phase, organizations recognize and interpret emerging problems or opportunities. The design (or solution generation) phase involves generating potential courses of action, and the choice (or solution selection) phase entails selecting among the developed alternatives.

While this model depicts decision-making as a structured yet recursive process, Ocasio (1997) argues that organizational decisions ultimately emerge from the allocation of managerial attention, shaped by broader organizational structures and contexts. This perspective, known as the attention-based view (ABV), conceptualizes decision-making as contingent on what decision-makers attend to, rather than on an exhaustive evaluation of available information. Extending this view, prior research suggests that decision outcomes are shaped not only by attention allocation but also by how decision-makers cognitively represent and frame underlying problems. Under conditions of ambiguity, decision-makers engage in sensemaking processes, constructing interpretations that render complex situations actionable.

Complementing these insights, Tversky and Kahneman (1974) advanced decision-making theory by empirically demonstrating that human judgment is systematically affected by cognitive biases and heuristics. Their work suggests that decision-makers cope with bounded rationality by relying on simplifying heuristics, leading them to satisfice rather than pursue fully rational, optimal outcomes. These insights extend to organizational decision-making, where choices are typically made under conditions of uncertainty and are therefore susceptible to bias and heuristic processing.

Decisions thus unfold under varying conditions and through multiple interacting processes. Simon (1987), for instance, distinguished between structured, semi-structured, and unstructured decisions, while Eisenhardt and Zbaracki (1992) characterized decision-making as an interplay of rational and political dynamics. Faced with competing forces and uncertainty, decision-makers increasingly rely on technologies to support and substantiate their reasoning. Yet, for decisions to yield observable outcomes, they must be implemented. Simon (1947) acknowledged this necessity by arguing that “factual judgements” require “the implementation of such goals” (p. 4), thereby implicitly pointing to implementation as a fourth, often underemphasized, stage in the decision-making process.

As organizations enact and adapt implemented decisions, they advance organizational learning by simultaneously exploiting established routines and exploring alternative courses of action. Exploitation of existing decisions enables organizations to capitalize on accumulated experience and reassess impacts on efficiency, whereas exploration of novel decisions is typically associated with greater risk-taking, agility, and experimentation. Risk-taking behavior, however, is shaped by performance feedback relative to currently salient organizational goals. Consequently, decision-makers often rely on short-term feedback—stimulating cognitive and factual responses—rather than anticipating distant future outcomes, as a means of containing uncertainty and avoiding risk.

This reliance on feedback renews managerial attention to performance gaps and introduces an implicit, yet recursive, phase in the decision-making process—feedback. By influencing the balance between exploration and exploitation, feedback affects the likelihood of organizational change and shapes organizational responses to discrepancies between realized performance and aspiration levels. Feedback, thus, operates not merely as an evaluative mechanism but as a central driver of adaptive decision-making dynamics.

Building on Mintzberg et al. (1976) initial, extensive catalog of decision-making processes, organizational decision-making unfolds within an environment characterized by bounded rationality, heuristics, biases, and the socio-material entanglement of human cognition with material and computational artifacts. To establish a coherent analytical foundation for this study, and drawing on the seminal contributions of the Carnegie Tradition, we conceptualize organizational decision-making as an iterative and recursive process comprising six interrelated components: attention, intelligence, design, choice, implementation, and feedback.

Thereby, the integrative framework captures decision-making as a dynamic cycle in which attention allocation initiates problem recognition, solution development and selection guide action, implementation enacts decisions, and feedback continuously reshapes subsequent attention, learning, and adaptation.

2.2 Decision support systems

Evolving from the early work of Scott-Morton (1971), decision support systems (DSS) are interactive computer-based systems designed to assist decision-makers in using models and data to solve unstructured problems. Subsequent scholarship broadened this initially narrow definition by emphasizing core DSS characteristics, including their focus on unstructured decision contexts, the integration of models and data, interactive usability for less technically skilled users, and adaptability to dynamic environments.

Through these features, DSS expand decision-makers’ cognitive and informational boundaries by enabling computational modeling, comparison, and justification of alternatives, while simultaneously shaping biases through structured data representa tions. As organizational decisions unfold across heterogeneous and evolving environments, decision-aiding technologies not only support better choices but also reshape organizational design, intelligence processes, and the conditions under which choices are made. In this sense, DSS reconfigure organizational decision-making by reallocating attention and cognitive effort across decision stages, while also altering conditions of coordination and accountability.

With the diffusion of computer-based DSS and intensified human-technology interaction, scholars increasingly questioned the relationship between the social and the material in decision-making. From a sociomaterial perspective, decision-making is enacted through sociomaterial practices rather than isolated technological artifacts. Here, the social (e.g., norms, rules, judgment) and the material (e.g., models, data, algorithms) are entangled, directing attention away from DSS as standalone tools toward the practices through which humans and technologies jointly perform organizational reality. Building on this relational view, recent scholarship reframes algorithmic agency not as a property residing in the algorithm itself but as an organizing capability that emerges through unfolding human-algorithm relations.

Consequently, DSS do not merely represent organizational contexts but actively participate in shaping them—most notably by redistributing decision rights, structuring attention beyond the provision of information alone, and reconfiguring what counts as legitimate evidence in organizational reasoning.

With the advent of AI and big data, information systems research increasingly shifted its focus from discrete system functionalities to the generation of insights from large-scale data, a development often described as one of the most consequential transformations in DSS history. In this vein, algorithmic technologies extended DSS toward augmented reasoning and decision-making. Operating through rule-based and statistical procedures, predictive AI accelerates the processing of vast information sets and engages predominantly with Kahneman’s (2011) System 2—a system of slow, deliberate, and analytical reasoning. This acceleration significantly reshapes human–machine relations in organizing and deciding, prompting scholars to conceptualize such configurations as “metahuman systems”, in which human cognition is augmented by computational intelligence.

Complementing this view, a systems perspective situates AI within broader networks of humans, artifacts, and routines rather than within an isolated human-AI dyad, drawing attention to the wider sociotechnical configurations through which decisions are produced. Accordingly, prior literature emphasizes the opportunities residing within AI-augmented decision-making, while outlining how human-AI coordination depends on emergent forms of hybrid cognitive alignment that calibrate the division of cognitive labor between human and machine.

However, predictive AI decisions remain grounded in predefined rules and procedures, lacking the capability to generate genuine novelty. While the outputs of

AI-driven decisions can be valuable, they continue to depend on human supervision, design, and programming. In contrast, intuitive and rapid decision processes—associated with System 1 reasoning —have traditionally been viewed as difficult to replicate computationally. Accordingly, AI has long been conceptualized primarily as a support tool and complement for analytical rather than intuitive decision-making.

Recent advances in GenAI, however, challenge this dichotomy. By producing novel outputs in open-ended tasks, GenAI applications may augment both System 1 and System 2 reasoning, extending metahuman systems with an entity capable of surfacing alternatives and insights that previously lay outside human attention.

2.3 GenAI in organizational decision-making

In contrast to predictive AI, GenAI does not merely perform and automate repetitive tasks based on rule-based systems but also undertakes tasks requiring creativity, decision-making capabilities, and interaction with human employees (van Heteren et al. 2024). Thereby, GenAI exhibits a qualitatively stronger form of algorithmic agency by constructing novel, context-specific outputs rather than merely selecting among predefined alternatives, influencing both the structure and substance of decision-making. From a relational perspective, GenAI should therefore be viewed not as a stable technological entity but as a phenomenon-in-the-making whose agency is enacted through the apparatuses (e.g., training data, prompting practices, deployment contexts) that materialize specific decision possibilities while foreclosing others.

Through its generative behavior and capabilities such as analogical, inductive, and causal reasoning, the integration of GenAI into decision-making processes elevates computational decision-making toward characteristics traditionally associated with biological systems. By compressing search, summarizing large datasets, and translating across technical domains, GenAI adoption not only yields productivity gains but may further reduce the bounded rationality of decision-makers. By surfacing insights from large and heterogeneous data sources, GenAI may reveal patterns that were previously undiscovered and reshape the information available during problem diagnosis.

These mechanisms jointly foreground three interrelated dynamics that distinguish GenAI-enabled decision-making: the socio-material entanglement of human judgement with generative computational practice, the redistribution of agency across human-AI configurations of conjoined action, and the emergence of hybrid human-AI cognition as a central coordination problem rather than a frictionless complement.

Therefore, value creation in this context depends on how humans and AI are coordinated. For instance, GenAI is already being used to support the evaluation of strategic decisions, generating strategic alternatives, and producing scenarios that allow decision-makers to explore second-order consequences prior to commitment. When paired with human oversight, GenAI may shorten the transition from sensing to seizing by accelerating the production of decision-ready artifacts. However, GenAI evaluations may vary across prompts, remain susceptible to hallucinations, and generate overly confident rationales. Thus, the stochastic nature of generative outputs necessitates verification, provenance, and accountability, rendering GenAI’s role in seizing highly dependent on process controls.

When firms complement GenAI with appropriate data assets and governance structures that prevent rhetorical coherence from substituting for validity, generative technologies may, ultimately, strengthen organizational dynamic capabilities.

The generative expansion of decision-making and decision-makers’ rationality, however, may both amplify and mediate biases and decision heuristics. Biases may enter decision-making processes through users’ prompting practices, biases embedded in training data, and decision-makers’ appreciation of or aversion to algorithmic recommendations. Consequently, biases in model outputs and biases in humans’ reliance on those outputs may jointly shape final decision outcomes. In this sense, GenAI may reproduce canonical cognitive biases and even amplify existing human biases, raising critical questions about the notion of positively connotated augmentation and the assumption of unbiased GenAI-supported decision-making.

Even in scenarios where generative models are comparatively unbiased, decision-makers may overweight their outputs due to fluency, confidence, or perceived authority, thereby triggering classic automation-bias dynamics.

Given the growing penetration of GenAI in decision-making contexts, scholars increasingly raise ethical concerns about its organizational use. Anthony (2021), for instance, highlight the black-box nature of GenAI as a barrier to decision-making, questioning transparency, trustworthiness, and fairness. In judgment-based environments, algorithmic inputs can intensify epistemic opacity and incur legitimacy penalties, requiring procedural safeguards. Moreover, inconsistencies in GenAI outputs further challenge the adequacy of GenAI in organizational decision-making, raising immediate questions of responsibility when generative systems emphasize or omit information.

To translate these concerns into AI governance, decision-makers must govern both the generative systems across their lifecycle and the decision processes into which they are embedded. To realize the benefits of GenAI in organizational decision-making, organizations should govern GenAI through structural (e.g., decision rights, accountability chains), relational (e.g., calibrated reliance, norms against “silent automation”), and procedural (e.g., impact assessments, human-in-the-loop) practices. Rather than treating GenAI as a purely technical rollout, it must be managed as an adaptive system co-evolving with organizational practices and the wider social environment.

2.4 GenAI and organizational decision-making—A status quo of recent SLRs

To contextualize our research, we reviewed the current state of systematic literature reviews examining the incorporation and influence of GenAI applications in organizational decision-making. To the best of our knowledge, only one SLR to date has explicitly addressed the interplay of generative applications and organizational decision-making. Yet, Porkodi and Cedro (2025) focus specifically on human resource (HR) decision-making, thereby limiting the scope of their review to a single functional domain. To broaden our understanding of the field, we therefore expanded the status quo assessment to include artificial intelligence more generally, which allowed us to identify three additional relevant SLRs. The methodological approaches, key contributions, and identified research gaps of these reviews are summarized in Tables 1 and 2 respectively.

Although all identified reviews follow SLR logic, they differ in their methodological rigor and inclusion criteria. Pietronudo et al. (2022) apply the structured approach proposed by Kraus et al. (2020), relying on clearly defined, quantifiable exclusion criteria, whereas Hani et al. (2024) follow the PRISMA framework outlined by Page et al. (2021a). In contrast, Porkodi and Cedro (2025) as well as Trunk et al. (2020) primarily justify article exclusion based on thematic relevance. With the exception of Hani et al. (2024), who identify 118 relevant articles, the remaining reviews rely on comparably smaller samples, which fall below the threshold of 100 ± 40 articles suggested by Marzi et al. (2025) as appropriate for conducting a comprehensive systematic review.

From a conceptual perspective, the identified SLRs are complementary rather than overlapping, each offering distinct contributions to the literature (see Table 2). While Pietronudo et al. (2022) and Porkodi and Cedro (2025) investigate the role of artificial technologies in decision-making at the departmental level, focusing on innovation and HR management respectively, Trunk et al. (2020) and Hani et al. (2024) adopt a broader organizational and industry-level perspective. Across these reviews, AI is consistently portrayed as enhancing organizational decision-making efficiency by supporting a variety of decision-related tasks.

Collectively, the SLRs identify persistent research gaps concerning the division of roles and tasks between humans and machines across the organizational decision-making process and call for the development of more comprehensive and integrated theoretical frameworks.

3 Methodology

To produce meaningful insights on a comprehensive understanding of the tasks and roles performed by GenAI applications in organizational decision-making, this study conducts a systematic literature review following the methodological approach of Tranfield et al. (2003). Acknowledging the evolving nature of literature reviews and given the novelty of the technology, this method is well suited to capturing the state of the literature that contextualizes the integration of GenAI in supporting decisions in organizational settings. By ensuring transparency, validity, reliability, comprehensiveness, and reflective

Table 1 Overview of SLRs on AI-supported organizational decision-making

Table 1 (continued)

Table 2 Key contributions and research gaps of previous SLRs agency, this approach reduces biases and errors, and fosters a deeper understanding on the use of generative tools for organizational decision-making. To enhance transparency and ensure replicability as well as intercoder reliability, two researchers conducted the SLR collaboratively, following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement approach. In addition, the approach was underpinned by inclusion and exclusion criteria represented in the work of Kraus et al. (2022). Accordingly, the SLR follows a three-step structure: data identification, data screening and inclusion, and data analysis and synthesis.

3.1 Data identification

To identify relevant articles, both authors independently, and subsequently jointly, discussed as well as elaborated on the development of the data funnel and criteria depicted in Fig. 1. Building on the work of Mariani and Dwivedi (2024), we utilized the search strings, including Boolean operators and field restrictions, ‘Generative Artificial Intelligence’ OR ‘GenAI’ OR ‘GAI’ OR GPT OR ‘Generative Pre-Training Transformer’ OR ‘ChatGPT’ OR ‘Copilot’ OR ‘Gemini’ OR ‘BARD’ OR ‘Perplexity’ OR ‘Large Language Model’ OR ‘LLM’ OR ‘Generative Model’ OR ‘Generative Network’ OR ‘AIGC’ OR Artificial Intelligence Generated Content’ OR ‘Claude’ AND Decid OR Decisi for a title, abstract, and keyword search in the databases EBSCO Discovery Services and Web of Science. A complete database-specific query syntax can be viewed in Appendix A.

To reduce multiplicity of duplications, the search has been limited to these two repositories as both databases belong to the largest scientific databases of academic articles. Following similar studies, the search was restricted to peer-reviewed journal articles (including both conceptual and empirical studies) with full-text availability, published in English, and covering the period from November 1, 2022, to March 17, 2025. In addition, letters were excluded, as this format typically does not provide sufficient methodological and contextual detail for systematic coding and is not peer-reviewed.

The selection of the time span is linked to the widespread adoption of GenAI following the release of ChatGPT in November 2022. This launch marked a qualitative shift in AI adoption, characterized by the mainstreaming of natural-language, general-purpose GenAI and an expansion in use-case breadth. Consequently, the focal mechanisms move beyond predictive analytics toward conversational sensemaking, rapid alternative generation, and novel human-AI delegation patterns in decision-making. Moreover, the generative capabilities of these systems enable an expansion of decision-makers’ bounded rationality, giving rise to new decision phenomena that justify a bounded search period commencing in November 2022.

The initial search led to an identification of n = 4438 articles (EBSCO Discovery Services: n = 2485; Web of Science: n = 1953), which was reduced to n = 2859 after the removal of duplications (n = 1579 articles removed). Following an objective coarse filter as first exclusion stage, a journal ranking exclusion (only articles to be consid ered following AJG ≥ 2 or IF ≥ 1.5; only one of the two conditions had to be fulfilled; n = 639 articles were below the ranking thresholds and subsequently removed), as proposed by Kraus et al. (2022) and Antonio and Kanbach (2023), has been applied, reducing the identified data sample to n = 2220 articles before screening.

3.2 Data screening and inclusion

The screening, inclusion, and exclusion of articles followed three iterations: title screening, abstract screening, and full paper screening. To ensure analytical rigor, the screening was conducted collaboratively by the two authors. Both authors initially and independently screened a subsample of the dataset to identify preliminary thematic inclusion and exclusion criteria, as well as interpretations. This approach served to assess whether the qualitative criteria were clearly interpretable or systematically ambiguous and to guard against post-hoc tightening or loosening of criteria to fit a preferred narrative. Subsequently, the two authors compared and aligned the qualitative criteria and jointly discussed thematic overlaps and differences.

Based on this alignment, the authors iteratively agreed-upon respective thematic exclusion criteria (as follows) and continued to conduct the screening collaboratively, thereby enhancing interrater reliability. After each screening stage (title, abstract, full paper), the authors aligned on the resulting data sample to be carried forward.

To ensure an appropriate balance between depth and breadth in article selection, and in line with prior reviews by Antonio and Kanbach (2023) and Nguyen and Kanbach (2024), articles focusing on decision-making in education, healthcare, and government-owned organizations were excluded. Although prominent in the pre-screened data sample, these three focus areas were excluded because decisions within these environments are often distributed through shared governance, constrained by strong norms of evidence-based practice, or embedded within multiple-principal environments. In contrast, our study focuses on organizational decision-making oriented toward firm performance and adaptation, operating under conditions of centralized accountability and high velocity.

In addition, to further ensure the validity of the SLR and its applicability to the derived research questions, articles that do not focus on a decision-making process as a whole, or that fall outside the corporate context, were excluded. In doing so, we strengthen the internal consistency of the review by focusing on decision-making processes solely within private companies and thus directly addressing the proposed research questions. In accordance with these thematic exclusion criteria, the screening process resulted in n = 920 articles before abstract screening (n = 1300 articles removed after title screening), n = 124 articles before full paper screening (n = 796 articles removed after abstract screening), and n = 60 articles before snowballing (n = 64 articles removed after full paper screening).

Finally, to ensure comprehensiveness and a full breadth of the SLR, snowballing was applied and resulted in the manual addition of n = 8 articles to the data sample. This technique is particularly useful as it helps to recover relevant studies that may have been systematically overlooked in database searches. In doing so, it addresses semantic disconnects where papers do not use the predefined search terms in expected combinations, yet still provide valuable contributions to the focal phenomenon. Accordingly, the application of snowballing strengthens both the coverage and saturation of the SLR.

Aligning with the overall aim of the study, the inclusion and exclusion criteria led to a comprehensive, final data sample of n = 68 articles, being in line with Marzi et al. (2025) threshold for conducting an appropriate SLR. A detailed overview of exclusions per iteration stage can be viewed in Fig. 1. Further, we provide a thematic inclusion rationale for all included articles in Appendix B.

3.3 Data analysis and synthesis

Following comparable studies, the data analysis employed an inductive approach informed by Glaser and Strauss (2017) and Gioia et al. (2013). To ensure relevant theory discovery and investigator triangulation, the research team collaboratively and iteratively coded the identified articles with a focus on the tasks GenAI performs in organizational decision-making.

Consistent with the screening procedure, the author team initiated the coding process with a pilot phase based on a smaller subsample. This step reduced the risk of silent divergence and facilitated the early development of a shared codebook. Subsequently, recognizing that coding behavior may vary across researchers, the authors continued coding independently and resolved divergent interpretations through consensus-building discussions. This process enabled the identification of underlying patterns of similarity and difference and strengthened the synthesis of the content analysis in terms of coding accuracy and inter-coder reliability.

In place of quantitative inter-coder reliability metrics, the coding process therefore drew on an interpretive orientation grounded in the conventions of inductive qualitative research, where shared conceptual understanding takes precedence over statistical concordance.

Overall, this procedure further ensured analytical distance and leveraged the benefits of multiple perspectives. Through this process, 53 first-order concepts were identified and subsequently assigned precise lexical descriptions. Next, relationships among these first-order concepts were examined and harmonized into 18 s-order themes, which were then organized into seven aggregated dimensions. This multistep aggregation enabled the analysis of concepts and themes from multiple perspectives and beyond their surface-level meaning, thereby uncovering less obvious patterns.

While first-order concepts and second-order themes capture the specific tasks GenAI performs in organizational decision-making, the aggregated dimensions reflect the broader roles assumed by GenAI, see Fig. 2. This analytical procedure supports theory development that extends beyond the existing status quo. To ensure the reliability and validity of the derived themes, newly formed categories were continuously compared with previously identified ones in line with constant comparison principles.

Notes. 1st order concepts and 2nd order themes reflect tasks GenAI performs in organizational decision-making; aggregated dimensions reflect roles GenAI upholds in organizational decision-making.

4 Results 4.1 Descriptive results

Overall, 68 relevant articles were identified that examine the tasks and roles of GenAI in organizational decision-making, with a noticeable increase in publications beginning in April 2024. The fragmented distribution of journals reflects the growing integration of GenAI into decision-making across diverse industries and organizational functions. At the same time, this dispersion underscores the absence of a unified and comprehensive understanding of GenAI’s tasks and roles in organizational decision-making as a whole.

This observation is further supported by the wide range of industries and departments addressed in the reviewed studies. Early evidence suggests that GenAI-enabled decision support is most prominently applied in supply chain management, human resource management, finance, and tourism. These domains appear to serve as initial focal points for exploring how generative technologies augment or reshape decision-making practices.

From a regional perspective, nearly three-quarters of the identified studies (approximately 73.5%) do not restrict their empirical or conceptual focus to a single country or region but instead adopt a global or cross-contextual perspective. This finding reinforces the view that GenAI’s influence on organizational decision-making is broadly pervasive rather than geographically selective. Nevertheless, a subset of studies (13.2%) concentrates explicitly on one of the three leading economic powers—China, India, or the United States—indicating emerging regional concentrations of GenAI-related research.

Regarding the level of analysis, 45.6% of the reviewed publications examine GenAI tasks and roles in decision-making while simultaneously considering their implications at both the individual and organizational levels. This dual-level focus highlights the intertwined nature of human and organizational dynamics in GenAI-supported decision processes. A detailed overview of these descriptive results is provided in Appendix C.

4.2 Thematic results

Following the systematic review of the relevant literature, 53 distinct tasks performed by GenAI in organizational decision-making were identified. These tasks were subsequently consolidated into 18 aggregated task categories, which together give rise to six distinctive GenAI roles: strategic analyst, automation specialist, futurist, process optimizer, human resource manager, and communicator. In addition, the review reveals a governing and collaborative human-AI role within GenAI-supported organizational decision-making, conceptualized as the AI-aware strategist.

The following sub-sections outline and substantiate each identified role by providing contextual explanations and illustrative examples drawn from the reviewed literature. All roles, along with their respective applications across the components of organizational decision-making, are conceptually illustrated in Fig. 3. A comprehensive overview of the identified GenAI tasks and their allocation to roles is provided in Appendix D.

4.2.1 GenAI as strategic analyst

Acting as a strategic analyst, GenAI performs three core task categories: analyzing data, identifying risk and opportunities, and generating alternative solutions. Closely linked to the automation of repetitive analytical tasks, GenAI substantially enhances the analysis of large and heterogeneous data sets. Its analytical capabilities span a wide range of organizational data, including employee performance metrics, historical sales data, customer feedback, trend analysis, supply chain data, and hospitality-related information. Across these contexts, GenAI accelerates the analysis of both structured and unstructured data. This analytical ease enables the generation of context-specific insights that support a broad set of organizational activities, including policymaking, finance, recruitment, and real-time support.

Building on large-scale data analysis, GenAI unlocks additional dimensions for the identification of risks and opportunities. Through activities such as risk-benefit analyses, viability assessments of large-scale projects, and risk mitigation via data synthesis, GenAI supports problem understanding and the development of mitigation strategies. By tracking issues (G. Chen et al. 2025a) and identifying root causes, detecting cybersecurity threats, and uncovering fraudulent activities, GenAI further contributes to emergency management and response capabilities. Beyond risk identification, generative tools also surface opportunities. When trained on organizational data, GenAI can identify growth potential in emerging and new markets, suggest new revenue streams for existing or novel products, and detect opportunities related to talent identification or infrastructure development needs.

Equipped with insights derived from data analysis and risk-opportunity screening, GenAI further supports the generation of alternative solutions. Drawing on its ability to propose options beyond existing knowledge structures, and to engage in structured brainstorming of potential outcomes, GenAI can generate strategic alternatives tailored to specific problems or preferences. Generative tools provide strategic recommendations regarding market segments to target or products to develop (G. Chen et al. 2025a), and may even suggest the most viable solution from an AI-based evaluative perspective. Once a set of plausible alternatives has been identified, GenAI can further assist decision-makers by simulating different scenarios and assessing the implications of each option from multiple analytical perspectives.

4.2.2 GenAI as automation specialist

As automation specialist, GenAI tasks are categorized in four main groups: automating repetitive tasks, automating content generation, supporting knowledge management systems, and executing processes autonomously. Focusing on task automation, GenAI performs a wide range of repetitive and clearly defined activities that typically require substantial human labor. In some use cases, GenAI demonstrates operational automation potentials of approximately 20%, thereby freeing employees to concentrate on tasks that demand higher cognitive capabilities.

For example, GenAI can function as a meeting minute tracker by transcribing verbal interactions into structured format (G. Chen et al. 2025a; Khan et al. 2024), schedule meetings, and automate inventory management, invoicing, and purchase orders in supply chain management. Moreover, it can detect faults in process system engineering, facilitate the posting of jobs, track recruitment outcomes and key performance indicators, automate the identification and extraction of relevant market information, and support code generation, debugging, and documentation. While this list is not exhaustive, these automation use cases are unified by their shared objective of reducing human workload, providing assistance, and enhancing efficiency.

A similar objective underpins the autonomous content generation capabilities of GenAI. Generative tools are increasingly used to create design proposals and product concepts, generate job descriptions, advertisements, and training material, produce problem-to-solution reports, summaries, strategic documents, financial reports, and updated manuals and standards.

Beyond task and content automation, GenAI advances organizational knowledge management systems by extracting and synthesizing knowledge from reports and products such as insights derived from sustainability reports or product features. Moreover, generative tools facilitate information retrieval from diverse databases or industrial process failure logs. Automated information gathering and retrieval not only enhance but also transform how organizational knowledge is stored, accessed, and distributed.

Equipped with comprehensive organizational information and enabled by continuous monitoring of financial and operational indicators, fault detection mechanisms, and response models, GenAI tools can further progress toward autonomous process execution. Such applications are already observable in logistics and execution systems, where GenAI reduces reliance on manual operations. To date, however, fully automated processes—including autonomous decision-making—remain largely confined to rule-based and tightly bounded operational environments.

4.2.3 GenAI as futurist

The breadth of analyses and assessments enabled by GenAI allows it to assume the role of a futurist, generating foresights and predictions while supporting innovation and exploration. In this role, GenAI conducts sentiment analysis in markets or risk management contexts, forecasts demand in supply chains as well as shifts in customer demands, and predicts market changes, such as movements in prices or evolving markets. Through these capabilities, GenAI facilitates the creation of alternative market scenarios, thereby equipping decision-makers with situational and forward-looking knowledge.

These insights support the exploration and identification of patterns and trends across customer segments, customer behavior, and broader future developments (J. Li et al. 2024a), enabling organizations to proactively respond to observed market dynamics. By streamlining foresight-related activities, GenAI contributes to smoother development processes and inspires decision-makers to identify and generate new product ideas, as well as to evaluate and select ideas based on anticipated competitive advantages.

In sum, by synthesizing predictive, interpretive, and generative capabilities, GenAI enables employees to peek into the future, supporting the assessment of both the feasibility and adequateness of organizational decision-making under conditions of uncertainty.

4.2.4 GenAI as process optimizer

In line with the automation of tasks and processes, GenAI tools not only execute but also optimize organizational processes, giving rise to two interrelated task categories within the role of GenAI as a process optimizer: optimizing organizational processes and improving organizational agility. Generative technologies possess the capability to move beyond established routines by identifying latent patterns in existing structures and autonomously adjusting processes accordingly. In certain use cases, GenAI demonstrates cost reduction potentials ranging from 25% to 60% and operational accuracy improvements of approximately 15%.

Such generative adjustments have already materialized across a wide range of organizational processes, including data-driven recruitment, optimal price negotiations, architectural design and land use planning, or the simplification of complex tasks. By transforming organizational value chains, GenAI’s optimizing capabilities contribute to accelerated workflows and increased process efficiency. Moreover, the integration of GenAI reduces the likelihood of both manual and cognitive errors, while decreasing the time spent on operational tasks. Collectively, these effects result in cost reductions across the organizational value chain.

Beyond efficiency gains, streamlined and cost-optimized processes enable more fluid and adaptable organizational structures, fostering more versatile and adaptive decision-making. By reacting dynamically to decision adjustments in complex and rapidly changing environments, GenAI strengthens organizational responsiveness and thereby enhances organizational agility overall.

4.2.5 GenAI as human resource manager

GenAI’s ability to provide neutral feedback and information in performance appraisals, candidate assessment, and evaluations grounded in rationale-based and objective criteria, enables it to assume functions traditionally associated with human resource management. In doing so, GenAI may foster greater transparency and perceived neutrality while simultaneously optimizing HR-related processes. By supporting the identification of cognitive biases and discriminatory patterns embedded in data, GenAI can further contribute to enhanced information transparency and data governance, affecting both employee-related and organizational data practices.

Across HR processes, GenAI supports the optimization of resource allocation, job and task categorization, and skill development. Empirical evidence suggests that the use of generative models can improve HR training outcomes by approximately 14%. By identifying resource inefficiencies and time constrains, GenAI enables the redesign and enhancement of HR processes through responsibility assessments and informed resource reallocation. In parallel, generative tools demonstrate expertise in mapping task interdependencies and identifying tasks with high potential for automation or augmentation.

To strengthen the skills and knowledge base of the existing workforce, GenAI further supports the identification of skill gaps, the creation of personalized and interactive training programs, and the development of analytical, critical, and creative thinking capabilities. However, realizing these benefits ultimately depends on employees’ willingness to engage in continuous learning and skill development, which remains a critical boundary condition for GenAI-enabled HR optimization.

4.2.6 GenAI as communicator

Fostering both internal and external communication, GenAI can assume the role of a communicator within organizations. Internally, GenAI can reduce information asymmetries by minimizing cognitive gaps between departments (Saxena and Rishi 2025;

Zhu et al. 2024), balancing knowledge disparities between SMEs and larger corporations, generating knowledge from both structured and unstructured data, and filling gaps in organizational knowledge repositories. By increasing the availability of information and access to diverse sources, GenAI may exert democratizing effects on decision-making processes, thereby improving internal communication.

In addition, GenAI facilitates the communication of managerial decisions and supports interaction between domain experts and data scientists, contributing to a shared organizational understanding. More dynamic knowledge interfaces, alongside increased interpretability and readability of complex legal or accounting documents, further strengthen this effect. Consequently, GenAI can support organizational discussions and engagement, guide employees through complex processes and concepts, and provide structured justifications for decisions made.

Externally, GenAI enhances organizational communication by increasing responsiveness to customers and markets, particularly in customer service contexts, while ensuring consistency of communication across multiple channels and situations. The customization and personalization of content further increase customer receptiveness to product recommendations and customer interactions more broadly. Finally, by enabling emotionally resonant communication, GenAI strengthens engagement with external stakeholders such as job applicants, tenants, and urban planners.

4.2.7 The AI-aware strategist

In contrast to the six aggregated GenAI roles discussed above, the role of the AI-aware strategist explicitly requires human intervention and positions humans as collaborative and governing actors in GenAI-supported organizational decision-making. Spanning three interrelated task categories— acknowledging GenAI limitations, enhancing cognitive capabilities, and governing with human oversight—this role is designed to mitigate the risk of overreliance on generative applications and generative authority. Prior research highlights multiple limitations associated with GenAI integration, including threats to decision quality, the erosion of diverse perspectives, biased outputs and insufficient explainability, as well as factual inaccuracies and hallucinations.

Additional concerns relate to output consistency and appropriateness, such as the legal validity of administrative decisions or the oversimplification of complex accounting judgments. Collectively, these limitations underscore the lack of fully autonomous GenAI decision-making in organizational contexts.

Despite these constraints, decision-makers increasingly recognize GenAI’s augmentative potential. Empirical studies point to improvements in problem-solving capabilities, mitigation of human cognitive limitations and heuristic biases, and greater diversity in the data and perspectives considered during decision-making. By fostering a form of human-AI co-intelligence, GenAI enables access to information and patterns that may otherwise remain outside human awareness. Moreover, by automating or accelerating routine cognitive labor, GenAI can free up time that decision-makers may reinvest in strategic tasks requiring judgment, sensemaking, and complex reasoning.

However, realizing these benefits requires a processual and iterative human-in-the-loop integration. Human oversight is necessary to critically evaluate generated content, intervene in or modify AI-supported decisions, and ultimately legitimize outcomes in organizational and societal terms. Given ongoing questions surrounding the authority and accountability of GenAI-supported decisions, final decision rights must remain with a human supervisor. Accordingly, the AI-aware strategist role emphasizes continuous human oversight to ensure the appropriateness, accuracy, and contextual fit of generative outputs, while aligning their use with ethical guidelines and broader societal values.

5 Discussion and framework development

In this study, we propose that GenAI’s tasks and roles can be mapped onto the six components of organizational decision-making, leading to the derivation of a GenAI task-role typology as represented in Fig. 3. Moving beyond this typological mapping, we argue that GenAI fundamentally reshapes and “rethinks” how organizational decision-making is conducted, thereby carrying transformative implications for the decision-making process (see Fig. 4). In doing so, we aim to enable a more nuanced view of the interplay between human cognition and algorithmic decision-making within organizations.

By identifying risks and opportunities, GenAI may function as an active attentional intermediary that summarizes, frames, ranks, and drafts interpretations. In doing so, it alters what becomes salient in the first place and, consequently, which issues attract managerial attention. By surfacing such risks and opportunities from unstructured cues, GenAI may reduce search and synthesis costs by addressing what Csaszar and Steinberger (2022) identify as the three core bottlenecks of organizational intelligence: the expansion of the search space, the enrichment of problem representations, and the acceleration of signal aggregation across distributed sources. The causal pathway thus runs from search-cost reduction to representation enrichment and, ultimately, to enhanced salience—not from data volume to decision quality.

This causal chain, however, may be conditional: it operates most effectively when decision-makers possess the domain expertise to interrogate and contest generative outputs, and it attenuates when organizations lack routines for reconciling machine-generated framings with experientially grounded judgment. Through clustering and prioritization, generative tools may therefore expand and accelerate the identification of emerging threats and nascent opportunities, and help decision-makers avoid single-frame lock-in during classification and sensemaking. Accordingly, the attentional intermediary role is most salient in environments characterized by high informational complexity, distributed signal sources, and time pressure, and least salient where decision problems are well-structured, signal sources are concentrated, and organizational experience with the domain is deep.

However, generative models may also overemphasize readily available or salient cues, crowding out quieter yet material risks and thereby evolving into a new attention structure with its own embedded biases (Y. Chen et al. 2025). Accelerated issue identification may therefore be accompanied by challenges related to accountability, reliability, and verification. Conversely, organizations that embed deliberate verification routines and maintain consistent oversight of generative outputs may be able to exploit GenAI’s attentional reach while containing its agenda-distorting effects. The critical boundary condition may therefore be the organizational design of the human-AI interface at the attention stage, rather than the technical capability of the generative model itself.

Moreover, a significant expansion of decision-makers’ bounded rationality can be observed in GenAI’s capability to analyze vast amounts of data and support knowledge management, lowering the costs of data triage, structuring, and interpretation. By enabling interaction with complex datasets through conversational prompts, GenAI introduces novel decision-support possibilities and extends versatility across intelligence activities. In this way, GenAI may support the development of a more systematically derived organiza tional knowledge base, allowing for faster decision cycles even among less technologically proficient decision-makers. The causal pathway through which this occurs may be grounded in the behavioral theory of the firm.

By lowering the cost of information retrieval and structuring, GenAI reduces the pressure on decision-makers to satisfice on the basis of readily available cues, thereby expanding the effective search radius before a decision is formulated. This mechanism is most potent when organizational decision problems require the synthesis of heterogeneous and large-scale datasets—conditions under which human working memory is a binding constraint. It attenuates in decision contexts that are time-compressed or highly tacit, where heuristic judgment reflects accumulated relational expertise that machine cannot, as yet, replicate. However, when GenAI-generated content is reintegrated into organizational knowledge bases without adequate controls, errors may compound over time.

This risk intensifies in proportion to the degree of organizational reliance on GenAI-generated content as primary input, indicating a feedback dynamic that has no parallel in prior DSS theory. The boundary condition moderating this contamination risk may be the organization’s data governance architecture: firms that maintain clear provenance tracking, version control, and human editorial authority over knowledge base inputs can interrupt the compounding dynamic.

Mindful of its limitations, generative tools enable decision-makers to design decision solutions based on vast, structured datasets and synthesized insights. By automating content generation and iterating alternative solutions, GenAI supports foresight development, prediction, and the exploration of innovation during the design phase. In particular, GenAI can rapidly draft decision artifacts central to design work, such as problem statements, option descriptions, business cases, and implementation outlines. This is especially relevant because design activities are often bottlenecked by translation work across domains and stakeholder groups.

By increasing both the diversity of options and the speed of their generation, GenAI can expand the pool of considered alternatives and stimulate decision-makers’ creativity. In doing so, GenAI not only lowers the entry barriers to creative work but also enables the simulation of trends based on historical data and market patterns. However, uncoordinated GenAI use entails the risk of homogenizing alternatives, as generative models may gravitate toward high-probability patterns embedded in their training data. When such models systematically overrepresent “typical” solutions, organizational exploration may become biased, crowding out genuinely novel approaches.

The causal mechanism underlying this homogenization is theoretically grounded: generative models are trained to produce statistically likely outputs, which implies that organizational exploration conducted primarily through GenAI will converge on solutions that were plausible given past data rather than genuinely novel given present conditions. This is precisely the exploitation bias that March (1991) identified as the structural risk of organizational learning—except that GenAI introduces it at the option-generation stage rather than at the routine-selection stage.

By combining factual knowledge with creative reasoning, GenAI’s benefits extend beyond supporting the initiation of innovation during the design stage to influencing the degree of idea variation and the timing of idea generation.

When it comes to assessing the validity of generated ideas and selecting among alternatives, however, GenAI’s influence on the choice stage remains limited.

Fully automated AI-enabled decisions are largely confined to operational domains, such as manufacturing, where decision contexts are clearly defined and closed. In contrast, strategic decision-making environments tend to favor human-AI augmentation, with humans retaining control over the final choice. The causal logic underlying the choice-phase gap may be reflected in the structural properties of the phase itself: choices, unlike predictions or designs, are irreversible commitments made under residual uncertainty, and their legitimacy depends on the accountability of the agent who commits. When agency is distributed across human-AI configurations, traditional accountability structures become difficult to assign, and stakeholders resist configurations that diffuse responsibility without diffusing risk.

This constitute a structural accountability threshold, not merely a psychological one linked to algorithmic aversion. The boundary conditions that modulate this threshold may therefore be rooted in the (ir)reversibility of decisions. Choices that are low-stakes, reversible, or operationally routine face a lower accountability threshold and are therefore more amenable to GenAI involvement. Where these characteristics are absent, decision-makers may limit the autonomy of generative models and display reservations toward accepting accountability for artificially generated choices.

Once decisions are made—regardless of whether they are human- or technology-based—GenAI can support their translation into concrete tasks, routines, assignments, timelines, and communication artifacts. By automating and accelerating repetitive and clearly specified activities, GenAI may function as an orchestration layer that enables decision-makers to leverage the extensive processing capabilities of generative applications. In doing so, GenAI facilitates the conversion of decisions into coordinated execution across roles, functions, and knowledge boundaries. By accelerating workflows and operational process improvements, GenAI can thus reduce implementation cycle times and variance in execution quality while simultaneously improving task-skill fit.

Causally, GenAI enables cognitive offloading by automating translation work across roles and functions, reducing working-memory demands placed on individual decision-makers, and enabling more consistent execution of the decision logic formulated at the choice stage. By accelerating the production of communication artifacts and role assignments, moreover, GenAI compresses the coordination lag that typically separates commitment from action in complex organizations.

AI-enabled implementation, however, is not without challenges. Concerns related to authenticity, reliability, and conformity to organizational standards accompany the use of generative outputs. Particularly in customer-facing contexts, AI-generated artifacts may deviate from established brand or organizational standards, posing reputational risks. These concerns underscore the necessity of human-led quality gates to ensure that outputs consistently meet organizational, ethical, and brand requirements.

In the feedback phase, GenAI primarily reshapes organizational decision-making by compressing sensemaking and communication work, thereby enhancing perceived transparency and neutrality while accelerating organizational learning cycles that under pin agility. By enabling faster and broader diffusion of feedback, GenAI may strengthen shared situational awareness across organizational units. This aligns with sensemaking research emphasizing that organizations continuously construct and revise interpretations through communicative processes. The causal pathway from GenAI-enabled feedback compression to enhanced organizational learning may be rooted in Greve's (1998) performance-feedback model. Because organizations learn when performance gaps are accurately registered and broadly communicated, GenAI’s capacity to compress sensemaking work may significantly accelerate these dynamics.

Accordingly, this mechanism may be most salient in organizations where feedback loops have historically been slow due to communication overhead or information asymmetry, and least salient where feedback loops are already fast and informal.

At the same time, GenAI may also emerge as a gatekeeper of organizational meaning. What is summarized, emphasized, or omitted can steer collective interpretations, particularly when generative summaries are treated as authoritative rather than as provisional and contestable accounts. GenAI may produce coherent and persuasive rationales that mask underlying uncertainty or misrepresent causal structures, thereby fostering a false sense of control and understanding. Accordingly, research on explainability cautions that transparency is multidimensional and must be deliberately aligned with organizational risk profiles and accountability requirements.

In addition, the professional and neutral tone of generative outputs may obscure contested assumptions, rendering feedback seemingly objective even when it embeds biases originating in training data or in the organization’s own historical records.

Mindful of the computational and human limitations inherent in decision-making, GenAI’s augmentative capabilities must be deliberately combined with judgment-based human capabilities to foster stakeholder alignment, calibrated reliance, and co-deciding. In line with Ivanov (2023), our framework proposes a human-in-the-loop configuration, which we term the AI-aware strategist. This collaborative human-AI role positions humans as the central and accountable actors in organizational decision-making, while emphasizing the augmentative potential of AI-enabled decision support systems as enacted through their users. Thereby, human oversight is paired with algorithmic support, and the transformative or “rethinking” potential of GenAI is largely shouldered by its human counterpart rather than by the technology itself.

We therefore argue that the beneficial effects of GenAI-supported organizational decision-making can only be realized when humans actively exercise control, supervision, and critical judgment over generative technologies. GenAI applications should thus be conceptualized and implemented as augmentation tools that enhance rather than replace human decision-making capabilities, supporting a reconfiguration of how organizational decision-making is performed.

6 Avenues for future research

This systematic literature review not only proposes a typology and process model of GenAI tasks and roles in organizational decision-making but also identifies contextual, theoretical, and methodological research directions (see Table 3). From a contextual perspective, the reviewed literature highlights the growing penetration of GenAI in organizational decision-making, with increasing application across tasks and processes. While our study documents the use of GenAI in decision-making across industries such as retail, renewable energy, and construction (G. Chen et al. 2025a), as well as across departments including human resource management, finance (X. Li et al. 2024), and marketing, it remains unclear how industry-and department-specific characteristics shape the integration of GenAI into decision-making processes.

As illustrated by studies in manufacturing contexts, clearly structured and formalized process flows enable GenAI to exercise higher degrees of autonomy, including participation in the choice phase of decision-making. Accordingly, future research could explore how variations in formal versus informal decision-making structures condition GenAI adoption and autonomy.

From a theoretical perspective, the identified GenAI tasks and roles imply substantial transformations across all components of organizational decision-making. As GenAI increasingly flags incidents and synthesizes information, it mediates managerial attention by shaping information availability and prioritization. Future research could therefore extend the attention-based view by examining AI-mediated attention, including algorithmic attention curation and organizational controls designed to prevent AI-induced attention narrowing. In the same vein, and noting that the novel socio-technological effects of GenAI differ from those of more traditional technologies, future research may further assess GenAI as a novel form of material artifacts shaping organizational attention.

In doing so, future studies may analyze the differential effects of GenAI on attention relative to those of more traditional material artifacts.

In the intelligence phase, GenAI shifts the emphasis from information gathering toward interpretation and evaluation of outputs. Future research may therefore explore how GenAI reshapes the concept of bounded rationality in organizational decision-making, given that information can be generated within seconds. This may prompt future discussion of whether and how GenAI and process automation may ease the path toward decision-making under perfect information provision.

In the design phase, the integration of GenAI may raise concerns related to a creativity-diversity tradeoff, as generative systems bias solution spaces toward high-probability patterns. Future research could therefore advance theories of human-AI collaboration by examining how GenAI use affects idea diversity and how homogenization effects might be mitigated. In doing so, such work would refine the discourse on GenAI tasks and roles within a more specific and nuanced domain of decision-making.

With respect to the choice phase, the scarcity of studies addressing GenAI-supported decision selection highlights unresolved tensions between automation and augmentation. Ambiguities surrounding responsibility and accountability continue to limit the deployment of fully autonomous systems. Accordingly, future research could contribute to the development of a theory of accountable autonomy that enables GenAI participation across the entire decision-making process.

In the implementation phase, organizations increasingly rely on GenAI for automation and process optimization. However, overreliance on artificial execution raises concerns related to implementation risk and accountability shifts. In particular, when strategic decisions are implemented and communicated by an artificial agent, employees’ receptiveness and the decisions’ effectiveness remain opaque. Therefore, future research may examine these emerging dynamics in strategic management, assessing how GenAI reshapes firm-specific implementation strategies and effectiveness.

Finally, in the feedback phase, GenAI may foster more personalized and neutral communication. However, GenAI-mediated summarization and sensemaking may centralize narratives and gatekeep the flow of information. Such mechanisms may thus constrain organizational learning and inhibit innovative initiatives within firms. Building on research on feedback, exploration, and exploitation, future studies could therefore investigate whether AI-mediated feedback systematically favors exploitation over exploration and how such tendencies might be counterbalanced.

From a methodological perspective, the review reveals a scarcity of experimental studies examining the effects of GenAI across the entire decision-making cycle. While existing research often analyzes selective decision stages, future studies could employ laboratory and field experiments to assess organizational outcomes resulting from end-to-end GenAI integration. Beyond outcome measurement, further research could theorize decision protocols governing GenAI use and examine organizational forms that enable or constrain generative decision-making.

7 Conclusion

GenAI holds the potential to disrupt how organizations make decisions at both the individual and enterprise level. However, the academic discourse on GenAI’s tasks and roles in organizational decision-making remains fragmented. Therefore, we conducted a systematic literature review and identified 68 relevant publications examining the integration of GenAI into organizational decision-making. Across these studies, we identified 53 distinct tasks and consolidated them into 18 task categories that GenAI performs throughout organizations’ recursive decision-making process. These task categories were further synthesized into six active roles and one collaborative role assumed by GenAI.

By mapping the identified tasks and roles onto the organizational decision-making components of attention, intelligence, design, choice, implementation, and feedback, we develop a conceptual framework that systematically structures GenAI’s potential to accelerate and enhance decision-making while emphasizing the continued necessity of human intervention across all stages. Notably, the choice phase of organizational decision-making remains largely unpenetrated by the processual capabilities of generative applications, indicating limited deployment dynamics and highlighting a critical boundary condition for GenAI integration.

7.1 Practical implications

While the bottleneck in the choice phase of organizational decision-making necessitates further research, emerging studies indicate a gradual progression of GenAI into the choice phase in specific industries, such as manufacturing. When decision-making processes follow clearly defined rules and operate within closed environments, GenAI can support the selection of optimal solutions. Decision-makers must therefore carefully assess the contextual characteristics of decision environments when considering the application of GenAI in organizational decision-making.

Prior to deploying GenAI, decisions may be classified along four dimensions, and the role of GenAI should be explicitly defined. First, decision-makers should assess the structural properties of the decision and determine whether it follows clear rules and stable evaluation criteria. Second, the decision environment should be evaluated, particularly whether it is closed or open, and whether ambiguity or external dependencies are present. Third, the scope of the decision must be considered, including its reversibility, potential safety or legal implications, and possible reputational consequences. Fourth, data availability and the feasibility of output verification should be assessed, and appropriate quality assurance mechanisms implemented. Decisions conducted in highly structured, closed environments with verifiable data and low stakes may be candidates for full automation.

In all other cases, GenAI should function as decision support rather than as the final authority.

Accordingly, industry- and decision-specific nuances must be considered. With the proposed typology and processual framework, we aim to provide clarity regarding the tasks and roles GenAI can assume in organizational decision-making and to offer a starting point for managerial discussions on where the deployment of GenAI is most beneficial. As GenAI’s capabilities to augment human decision-makers continue to improve, tasks characterized by clear structure are particularly amenable to automation and enhancement. However, beyond the automation of repetitive tasks and processes, decision contexts requiring novel insights and shifts in perspective also constitute a promising domain for GenAI integration.

Despite persistent aversion toward GenAI utilization, clear governance structures are required to delineate areas of application and preserve human authority. As questions surrounding the reliability and accountability of GenAI-supported decisions remain unresolved, humans must retain the ability to override GenAI recommendations. For example, managers may define conditions that trigger mandatory escalation, such as missing data, policy conflicts, or high-stakes outcomes, and require human oversight of generative outputs. At the same time, evidence from manufacturing contexts suggests an expanding range of decision situations in which GenAI-supported choices may yield superior outcomes. We therefore propose a selective application of GenAI roles and tasks within organizational decision-making, guided by the collaborative principle of the AI-aware strategist.

To this end, employees must develop skills that directly affect decision quality, including problem formulation, output verification, bias and overconfidence detection, and an understanding of contexts in which GenAI should not be used. By adhering to a human-in-the-loop principle, managers can thereby ensure high-quality decision-making and shared human-AI accountability.

7.2 Theoretical implications

Theoretically, this study makes three core contributions. First, it demonstrates how established theories of organizational decision-making require reassessment in light of GenAI-enabled decision processes (see Table 3). For instance, when generative models identify risks and opportunities, managerial attention may be systematically channeled toward high-probability patterns, potentially crowding out less salient yet equally consequential organizational challenges. Similarly, as GenAI expands organizational knowledge management and accelerates information availability, the applicability of bounded rationality warrants re-examination.

While bounded rationality remains present in phases dominated by information processing, search, and synthesis, the uneven distribution of task categories, combined with the near absence of GenAI from the choice phase, implies that it is alleviated only selectively. This invites a phase-contingent theory of AI augmentation, in which the cognitive benefits of GenAI are structurally bounded by the nature of each decision phase, not merely by task complexity or data availability.

Second, this study proposes a unified typology of GenAI tasks and roles in organizational decision-making (see Fig. 3). Although the typology is inherently static, it enables the identification of distinct task clusters and roles, thereby structuring and streamlining the academic discourse on algorithmic agency in decision-making. In doing so, the typology clarifies distinct contributions of GenAI agency within collaborative human-AI decision-making and highlights the growing and evolving influence of socio-materiality within organizations. Rather than merely priming human decision-making, our systematic review outlines how, when, and to what extent material artifacts can now autonomously support and facilitate distinct tasks within organizational decision-making.

Moreover, by identifying a collaborative human-AI role, we further outline how GenAI's procedural capabilities can be augmented by human cognition, and vice versa. Taken together, these insights may support on facilitating more targeted theory development, sharper empirical inquiry, and clearer practical implications for the discourse on socio-materiality, algorithmic agency, and human-AI hybrid cognition.

Third, we theorize how GenAI actively “rethinks” organizational decision-making by mapping its influence across the six recursive components of the decision-making process (see Fig. 4). This processual perspective not only illustrates where and how GenAI intervenes in decision-making but also provides a conceptual foundation for identifying phase-specific opportunities and challenges. In particular, the findings on the unpenetrated choice phase may form an argument that generative algorithms excel primarily in formal but cannot exercise substantive rationality. Thus, the choice phase of organizational decision-making may constitute a structural accountability threshold: a point in the decision process where the human-AI system cannot distribute responsibility further without undermining the organizational and ethical legitimacy of the decision itself.

7.3 Limitations

Alike other research, this study has some limitations. Although the literature review has been conducted following clear steps and criteria, as well as keywords being derived from established research, the selection of keywords might have inaugurated biases, influencing the search and assessment. To navigate the vast and rapidly advancing landscape of GenAI, however, we tried to navigate biases by incorporating the most prominent GenAI applications and key terminology. In addition, we have excluded non-peer-reviewed articles, such as conference proceedings, and focused solely on two databases, which might have led to the neglect of relevant and most recent research. Similar to other SLRs, our study focused on articles written in English, omitting research conducted in other languages.

Although following systematic article selection according to Tranfield et al. (2003) and Page et al. (2021a), the selection of articles and coding of keywords might have been subject to a researcher’s bias. This subjectivity should be considered when interpreting the derived task-categories and roles performed by GenAI in organizational decision-making. Therefore, future researchers are encouraged to elaborate on the insights uncovered in this systematic review, by analyzing and exploring further ways in which GenAI may influence the realm of organizational decision-making.

Appendix A

Sec Table 4.

Appendix B

Sec Table 5.

Appendix C

Sec Table 6.

Table 6 Sample characteristics

Table 6 (continued)

Appendix D

Sec Table 7.

Table 7 GenAI’s tasks and roles according to relevant literature

Table 7 (continued)

Table 7 (continued) data collection and analysis were performed by Niklas Schulte and Dominik K. Kanbach. The first draft of the manuscript was written by Niklas Schulte and Dominik K. Kanbach and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding Open Access funding enabled and organized by Projekt DEAL. The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Data availability Data sharing is not applicable.

Declarations

Competing interests The authors have no relevant financial or non-financial interests to disclose.

Author contributions All authors contributed to the study conception and design. Material preparation,