AI-Enabled Detection of Governance Dilemmas in Digital Transformation Projects: A Micro-Longitudinal Study of Corporate Innovation Incubation

1. Introduction

Digital transformation (DT) has become a strategic imperative for established organizations seeking to maintain competitive advantage in an era of rapid technological change [1,2]. A primary vehicle for delivering DT is via projects, where discrete, time-bound initiatives are launched to implement new technologies, develop new business models, and build new organizational capabilities [3,4,5,6]. Corporate innovation labs, startup incubators, and agile development teams are now common features of organizational landscape, practicing project-based approach to navigating the uncertainties of digital change [7,8,9].

However, the performance of these DT projects is notoriously challenging to manage. Traditional linear project management approaches often encounter limitations when attempting to cope with these dynamic contexts. Consequently, an increasing body of research has begun to investigate agile project management in facilitating digital transformation processes [10,11,12,13,14]. Agile originated in the field of software development, where its core principles emphasize iterative development, continuous feedback, close team collaboration, and a rapid response to change [12,14,15,16]. Within the broader project management literature, agile is commonly defined as an approach grounded in short iterative cycles, cross-functional teamwork, and ongoing stakeholder engagement, with the aim of flexibly adjusting project objectives and development pathways in response to complex and evolving environments [10,12,13,15,17]. With the continued advancement of digital technologies, agile practices have gradually extended beyond their original software development context to a wider range of digital transformation initiatives and are increasingly regarded as an important managerial approach for enabling organizational innovation and enhancing adaptability in dynamic market environments [13,14,18]. While agile and lean startup methodologies have provided frameworks for managing technical and market uncertainty at the team level [19], they often clash with the established governance structures of the parent organization [18,20,21]. Consequently, many promising DT projects stall not for lack of viable ideas or technical feasibility, but because they become ensnared in a web of bureaucratic hurdles, resource allocation conflicts, and decision-making bottlenecks [22]. This creates a critical challenge for project management: how to govern DT projects in a way that balances the need for speed and agility with the realities of corporate oversight and strategic alignment [23,24].

A growing body of research suggests that DT projects fail not primarily because of technical limitations or resource scarcity, but because of governance and decision-making bottlenecks [5,25,26]. Kulichyova et al. [27]. identify four macro-level restructuring dilemmas (i.e., structural, performance, technology, and leadership), that characterize digital transformation in large organizations, with the structural dilemma (balancing agility and stability, decentralization and centralization) emerging as the most pervasive. Despite this recognition, empirical evidence on where and how governance bottlenecks manifest over time at the micro level of day-to-day real project interactions remains scarce [28,29,30].

There is a pressing need to address this gap. If governance bottlenecks are structural and recurrent rather than episodic and idiosyncratic, then interventions targeting individual decisions or specific technical problems will be insufficient. What is needed is a systematic understanding of the types of governance dilemmas that recur in DT projects, the decision stages at which they concentrate, and the temporal patterns through which they persist or resolve. This governance tension has direct, though often unmeasured, implications for project performance. Decision latency at the seizing stage, as we will demonstrate, translates into tangible project costs, including schedule delays, rework, missed market windows, and increased coordination overhead as teams escalate issues through the organizational hierarchy [31,32]. Although this study does not directly quantify these costs, it proposes governance tension profile as a quantifiable leading indicator of project performance degradation, shifting the focus from lagging indicators (cost/schedule overruns) to proactive governance monitoring. Such understanding will enable project managers and DT leaders to design governance mechanisms, (such as stage-gate rules, decision rights, fast-path approval processes) that address the root causes of project stalling rather than its symptoms. This study directly addresses the Special Issue's themes of AI-enabled project management and data-driven decision making by offering an operationalizable meeting-analytics pipeline for governance monitoring in DT projects.

Process research in organizational theory has long argued that understanding phenomena requires attention to the temporal unfolding of events and decisions rather than cross-sectional snapshots [33,34]. Pettigrew [35] laid the foundation of longitudinal research on organizational change, emphasizing the importance of studying processes 'in context and over time.' More recently, scholars have called for micro-level process research that examines the episode-by-episode dynamics of strategic action [36,37], yet such research remains methodologically challenging due to the volume and granularity of data required.

This paper addresses this gap by conducting a micro-longitudinal study of project governance meetings in two corporate startup initiatives participating in the same internal incubation program. We develop and apply an AI-enabled analytical pipeline to systematically identify and analyze governance dilemmas as they emerge, evolve, and are navigated in real-time. Our research is guided by the following questions: RQ1. What are the key project governance dilemmas that emerge in the day-to-day interactions of corporate innovation incubation projects? RQ2. At which stages of the innovation process (sensing, seizing, reconfiguring) do these dilemmas concentrate? RQ3. How do these dilemmas evolve over time, and what are their temporal patterns?

By answering these questions, this study makes three major contributions. First, we provide a rich, empirically grounded typology of 13 project governance dilemmas, revealing a concentration of structural and seizing-related tensions that constitute a critical governance choke point in corporate DT projects. Second, we demonstrate the value of a micro-longitudinal, AI-assisted approach to studying project governance, offering a replicable method for detecting governance tensions from the unstructured data of meeting conversations. Third, we bridge the macro-level framework of organizational paradoxes [27,38] with the micro-level dynamics of project decision-making, proposing two meso-level dilemmas that act as transmission mechanisms.

2. Theoretical Background

2.1. Digital Transformation as Project-Based Organizing and Governance Work

Digital transformation (DT) initiatives are commonly realized through project-based organizing, where cross-functional teams coordinate technology, process redesign, and organizational change under high uncertainty and interdependence [39,40,41]. In this setting, coordination cannot rely solely on formal plans; decision-making becomes distributed, iterative, and continuously renegotiated as new information emerges and constraints shift [42,43]. Project governance therefore plays a central role in shaping how DT projects progress, because it determines how decisions are authorized, how trade-offs are adjudicated, and how accountability is allocated across sponsors, project teams, and functional stakeholders [5,6,44,45,46].

Project governance refers to the formal and informal arrangements through which decision rights, accountability, escalation paths, and control mechanisms are defined and enacted over the life of a project [47,48]. In digital transformation projects—where uncertainty is high and dependencies cut across functions—governance is not a static structure but a recurrent practice: it is performed through recurring forums (e.g., steering committee reviews, sprint demos, and stage-gate approvals) where teams negotiate priorities, authorize spending, resolve interdependencies, and secure sponsor buy-in [49,50]. Accordingly, we treat weekly governance meetings as a privileged empirical window into project governance, because they concentrate the 'moment of truth' interactions in which control and autonomy are balanced, compliance constraints are interpreted, and resource commitments are authorized or deferred [51].

2.2. Agile Project Management and Digital Transformation

Existing research has examined the role of agile project management in digital transformation from multiple perspectives. First, a stream of studies has focused on the process through which organizations transition from traditional project management approaches to agile methods. For instance, Burga et al. [12], through a case study of information systems development teams, found that organizations introducing agile practices often undergo several stages of transformation, including structural adjustments, enhanced team autonomy, and the reconfiguration of decision-making processes. These organizational changes were shown to significantly improve teams’ adaptability and innovative capacity.

Second, scholars have investigated how agile team structures and continuous delivery mechanisms influence the performance of digital projects. In large-scale digital initiatives, autonomous teams and continuous deployment mechanisms can substantially increase project responsiveness while reducing technological uncertainty [52].

In addition, some research analyze the impact of agile project management from governance and organizational management perspectives. For example, the agile governance theory [16] suggests that flexible decision-making structures combined with rapid feedback mechanisms enable organizations to achieve higher levels of project performance and organizational adaptability in complex project environments. Similarly, Koudriachov [53] emphasizes that effective team collaboration, the utilization of digital tools, and flexible process design constitute critical factors contributing to the success of agile projects.

2.3. Governance Dilemmas as Paradoxical Decision Situations

Research on organizational paradox and tensions emphasizes that competing demands often coexist and persist rather than being fully resolved [38,54]. Smith and Lewis [38] define paradox as 'contradictory yet interrelated elements that exist simultaneously and persist over time,' and argue that effective organizations develop the capacity to navigate rather than eliminate such tensions. In project settings—particularly DT projects—these competing demands are frequently expressed as governance dilemmas: decision situations in which two legitimate and actionable courses of action are simultaneously salient (e.g., autonomy vs. control; speed vs. compliance; experimentation vs. standardization) [55,56]]. Unlike one-off conflicts, governance dilemmas tend to recur because they reflect durable interdependencies and role expectations across organizational boundaries (e.g., parent organization vs. venture team; compliance functions vs. delivery teams; portfolio priorities vs. local learning) [23,57].

Kulichyova et al. [27] identify four macro-level restructuring dilemmas—structural, performance, technology, and leadership—that characterize DT in large established organizations. Their framework provides a valuable macro-level map of the tensions that DT programs face. However, it does not explain how these macro-level tensions manifest in the day-to-day micro-level interactions of project teams, nor does it account for the specific governance challenges of incubation-style corporate ventures. We build on their framework by examining how macro-level dilemmas are transmitted to and enacted at the micro level of project governance meetings [58].

We use the term 'governance tension' to denote an episode-level instance in which competing governance-relevant actions are articulated or enacted within a meeting interaction. We use 'governance dilemma' to denote a recurring trade-off pattern that aggregates multiple tensions across episodes into a stable, interpretable decision problem that project actors must repeatedly manage. This distinction between tension (micro, episodic) and dilemma (meso, recurring) is central to our analytical framework and follows the process-theoretic tradition of studying organizational phenomena at multiple levels of temporal granularity [34,35].

2.4. Dynamic Capabilities as a Process Lens for Decision Stages

To connect micro-level governance tensions to project progress, we draw on a process view of dynamic capabilities [59,60]. Teece [59] defines dynamic capabilities as 'the capacity of an organization to purposefully create, extend, or modify its resource base,' and decomposes them into three clusters: sensing, seizing, and reconfiguring. In a DT project context, sensing concerns how teams frame situations, identify opportunities or threats, and generate options for action [61]. Seizing concerns selecting among options and converting intentions into commitments—such as allocating resources, approving vendors, committing to roadmaps, and authorizing execution [62,63]. Reconfiguring concerns transforming routines, roles, and structures to sustain the new direction, including changes in governance arrangements themselves [64,65].

This process lens is useful because it highlights that decision work is not homogeneous: different kinds of constraints and tensions are likely to dominate at different stages, shaping how quickly projects move from ideas to execution [66]. Prior research has noted that established organizations often excel at sensing (they have sophisticated market intelligence and R&D functions) but struggle with seizing (converting insights into committed action) and reconfiguring (adapting legacy structures to support new directions) [59,67]. Our study tests whether this pattern holds at the micro level of project governance interactions.

2.5. Why Structural Dilemmas Become a Seizing Choke Point

We theorize that structural dilemmas become disproportionately visible at the seizing stage because seizing is the decision gate where abstract governance design choices are translated into binding commitments [68]. Structural tensions—such as autonomy versus control, compliance versus agility, and centralization versus decentralization—specify who is authorized to decide, what evidence counts as sufficient, which procedures must be followed, and how accountability is allocated [69,70]. These constraints may remain latent during sensing (idea generation and opportunity framing), but they become salient when projects must convert intentions into commitments: procure a resource, approve a vendor, allocate a budget, commit to a roadmap, or authorize execution [71]. As a result, structural dilemmas surface as decision latency and escalation cycles precisely at the point of resource-commitment (seizing), producing a systematic governance choke point that is decisional rather than ideational [72].

This theoretical mechanism has direct implications for project performance. Decision latency at the seizing stage translates into tangible project costs: schedule delays, rework when decisions are deferred and then made under greater time pressure, missed market windows, and increased coordination overhead as teams escalate issues through the organizational hierarchy [32,73]. Crucially, these costs are typically invisible to traditional project monitoring systems with pre-defined goals, which track lagging indicators such as cost and schedule variances but do not measure governance tensions that precedes them [10,18]. Our study addresses this blind spot by proposing governance dilemma frequency and persistence as leading indicators of project performance risk.

2.6. Conceptual Summary and Research Positioning

Taken together, this theoretical framework suggests that DT project performance may be shaped less by the availability of ideas and more by how governance arrangements channel decision rights and commitments under uncertainty [74]. If structural dilemmas dominate seizing, then recurring governance forums—where commitment decisions are made—should exhibit detectable patterns of tension concentration, persistence, and escalation. This conceptual framing motivates our micro-longitudinal, meeting-based approach and sets up our empirical focus on identifying and classifying governance dilemmas at the episode level. The framework also positions our study in relation to Kulichyova et al. [27]: while their work maps the macro-level dilemma landscape of DT, we descend to the micro-level to show how these dilemmas are enacted, concentrated, and temporally patterned in the day-to-day governance of corporate innovation projects.

3. Materials and Methods

3.1. Research Setting and Data

This study is based on a micro-longitudinal analysis of two corporate startup initiatives participating in the same internal incubation program within a large, multinational industrial firm (pseudonym: 'Corp'). The two initiatives, 'Startup 2' (a materials circularity service solution) and 'Startup 1' (a new battery charging service solution), were selected as representative cases of project-based DT initiatives operating with partial autonomy within a large parent organization. Both initiatives participated in the same structured incubation program, which mandated weekly pulse meetings as the primary forum for the startup team to review tasks, solve problems, and track progress against objectives.

The data corpus consists of the complete set of transcripts from 28 weekly pulse meetings held between November 2024 and April 2025. The meetings were conducted via Google Meet and involved the members of each startup team and a meeting facilitator (third author), following a consistent format of task review, problem-solving, and progress tracking against objectives. Transcripts were generated automatically using Tactiq, a real-time transcription plugin for Google Meet that produces time-stamped, speaker-attributed text. The raw transcripts were subsequently de-identified by the first author following the IRB-approved protocol: all participant names, client names, company names, and partner names were replaced with generic pseudonyms (e.g., P1, P2, [Client A], [Partner B]) prior to any analytical processing. The total dataset comprises over 500,000 words of transcribed text across the two initiatives.

This study was reviewed and approved by the Arizona State University Institutional Review Board (IRB ID: STUDY00023789; Expedited Review, Category (5) data/documents/records). The present article reports a secondary analysis of de-identified organizational meeting transcripts. A waiver of informed consent was approved because the materials were originally generated for operational documentation and were subsequently de-identified prior to research analysis. All participant names, client names, company names, and partner names were replaced with generic pseudonyms to protect confidentiality.

3.2. Analytical Pipeline

We developed a six-stage analytical pipeline to move from the raw transcripts to a theoretically grounded understanding of governance dilemmas (see Figure 1). This pipeline combines qualitative research principles with computational methods to ensure both interpretive depth and analytical rigor.

Stage 1: Contextual Reading and Episode Segmentation. The research team first conducted a thorough reading of all meeting transcripts to develop a holistic understanding of the context, key actors, and project trajectories. Following this, the transcripts were segmented into 711 discrete episodes, defined as bounded sequences of conversational turns unified by a common topic or decision focus [34,72,73], with the help of LLM and supervision of Author 1. No major segmentation errors were detected during checks and manual review. This segmentation provided the basic unit of analysis for the subsequent stages.

Stage 2: Tension Detection via LLM-Assisted Classification. The second stage involved systematic identification of decision-relevant governance tensions. We employed a hybrid approach combining rule-based linguistic screening with LLM-assisted semantic classification. First, a lexical algorithm screened the 711 episodes for linguistic markers of tension (e.g., contrastive connectives, modal hedges), reducing the pool to 187 candidate episodes. Second, a GPT-4-class LLM classifier, prompted with a precise definition of a governance tension ('a situation in which two legitimate but competing, actionable courses of action are simultaneously present'), was applied to each candidate episode. This process yielded 28 episodes containing high-confidence governance tensions. The LLM's classification accuracy was validated against human expert coding on a sample of 30 episodes, achieving a Cohen's κ of 0.82, indicating substantial agreement [75]. We prioritized precision over recall in this stage to minimize false positives; therefore, the results represent a conservative estimate of the total governance tensions present. Full prompt specifications and validation details are provided in Supplementary Materials.

Stage 3: Abductive Grouping into Governance Dilemmas. The 28 identified tensions were then grouped into broader governance dilemmas through an abductive analytical process [76,77]. This involved iteratively moving between the empirical data and existing theory to develop categories that were both empirically grounded and theoretically meaningful [78]. Author 1 independently grouped the tensions and then discussed with Author 2 and Author 3, resulting in a final set of 13 distinct governance dilemmas. The inter-rater reliability for this grouping task was high (Cohen's κ = 0.88).

Stage 4: Mapping to Dynamic Capabilities. Each of the 13 dilemmas was then mapped onto Teece's [59] framework of dynamic capabilities: sensing, seizing, and reconfiguring. LLM and Author 1 independently rated the relevance of each capability and assigned a primary capability. Agreement was 100%.

Stage 5: Longitudinal Analysis. To analyze the evolution of the dilemmas over time, we created a timeline visualizing the occurrence of each dilemma across the 28 meetings. This allowed us to identify temporal patterns, classified as: (a) single occurrence, (b) recurring, and (c) persistent (appearing frequently with intervals typically less than two weeks).

Stage 6: Synthesis of Meso-Level Dilemmas. In the final stage, we analyzed the relationships between our empirically derived micro-level dilemmas and the macro-level restructuring dilemmas identified by Kulichyova et al. [27]. This comparative analysis revealed clusters of micro-dilemmas that did not map cleanly onto a single macro-dilemma, leading to the synthesis of two meso-level dilemmas characteristic of the innovation incubation context.

3.3. Use of Generative AI

During the preparation of this study, the authors used a GPT-4-class large language model as an analytical aid for classification, as described in Section 2.2. The authors reviewed and validated all model outputs and took full responsibility for the content of this manuscript. GenAI was also used in planning and review phases of the paper (e.g. structuring and revision) The final version of the manuscript was written by humans.

4. Results

Our analysis revealed 13 distinct project governance dilemmas, which were unevenly distributed across the dynamic capabilities framework and exhibited distinct temporal patterns. The central finding is the identification of a governance choke point where structural dilemmas and seizing-related decisions intersect, creating a persistent source of tensions for the DT projects.

4.1. A Typology of 13 Project Governance Dilemmas

Table 1 presents the 13 governance dilemmas identified through our abductive grouping process. These dilemmas represent the recurring tensions that project teams and their sponsors had to navigate. They range from operational issues, such as balancing cost and quality (D06), to strategic choices, such as pursuing partnerships versus adapting the business model (D12).

To illustrate the nature of these dilemmas, consider the following three vignettes drawn directly from the pulse meeting transcripts. Each vignette is a verbatim excerpt (translated from Portuguese) from a specific meeting, with participants identified as P1–P4 to preserve anonymity.

Vignette 1: The Formalization Tension (D02 — Startup 2, Meeting 2024.11.11)

Context: The Startup 2 team discusses the need to move forward on a task but is constrained by a formal corporate process.

P3: "Before we move on, just a comment... I think it's important to formalize this. We've already moved ahead a bit, right?"P1: "[He] got excited there... but it's good. Now we have a problem on our hands to solve, a demand problem."P3: "No, but it's just to... we need to have the request formalized. We can't just go and do it."

This vignette captures a classic governance tension: the conflict between the parent company's need for standardized, auditable processes and the project's need for speed and agility. Both poles are legitimate—compliance is necessary, but so is speed. The dilemma is not about choosing one over the other, but about how to navigate the tension between them.

Vignette 2: The Partner Commitment (D01 — Startup 1, Meeting 2024.11.11)

Context: The Startup 1 team discusses how to engage a large potential partner who has been unresponsive.

P1: "Contact [Partner A] again, I think you had gotten the name, right? Contact them again, insist a little. But at the same time, let's look for other names, other companies that are similar."

P2: "I think it's worth trying one more time, but we need to think of a Plan B. We can't be dependent on them."

This second vignette illustrates the dilemma of Persistence with Difficult Partners vs. Seeking Alternatives (D01), a common seizing challenge where the team must decide whether to allocate more resources to a high-potential but difficult relationship or pivot to more accessible but potentially less valuable alternatives.

Vignette 3: The Pricing Model (D03 — Startup 1, Meeting 2024.11.18)Context: The Startup 1 team debates whether to offer a flexible, informal pricing model to test the market, or stick to a formal, compliant price point.P4: "What we tried to do was to keep the recharge price at a maximum of two reais to charge the guys."P1: "And I think you are discarding the half charge option too quickly. Haven't you tested it too?"P4: "Yes, so the ideal world would be to give a half charge, but what we..."P1: "It's because it's a test, right? Don't discard it so easily. You are the entrepreneurs, you have to take the leap of faith."

This third vignette exemplifies the Financial Incentives/Compliance vs. Market Realities dilemma (D03), where formal financial discipline clashes with the need for agile, market-driven experimentation.

4.2. The Seizing Choke Point: Concentration of Dilemmas in Decision-Making

When we mapped the 13 dilemmas to the dynamic capabilities framework, a striking pattern emerged (see Figure 2). An overwhelming majority of the dilemmas—9 out of 13 (69%)—were primarily related to seizing. These are dilemmas that arise at the point of decision-making and resource commitment. A further 3 dilemmas (23%) were primarily related to reconfiguring, involving the adaptation of operational and business models. Only one dilemma (8%) was primarily related to sensing, the identification of new opportunities.

This concentration at the seizing stage suggests that the primary challenge for these corporate innovation projects is not a lack of ideas or an inability to sense opportunities. Rather, the tension is concentrated at the moment of truth: when a decision must be made, resources must be committed, and the organization must move forward. Dilemmas such as D01 (Persistence with Difficult Partners vs. Seeking Alternatives), D04 (Growth Through Expansion vs. Focused Targeting), and D09 (Centralized Control vs. Team Autonomy) are all fundamentally seizing-related challenges.

4.3. Temporal Patterns: The Persistence of Structural Dilemmas

The longitudinal analysis revealed that the dilemmas were not just static categories but had distinct temporal signatures (see Figure 3). We found that 6 of the 13 dilemmas were single-occurrence events, tied to specific, non-repeating contextual challenges. However, 7 dilemmas were either recurring or persistent, indicating they were structural features of the governance landscape: 4 recurring dilemmas (e.g., D05: Thorough Investigation vs. Rapid Market Adaptation) and 3 persistent dilemmas (D01, D02, and D03) that reappeared frequently, often within 1–2 weeks, indicating unresolved underlying tensions.

Importantly, the dilemmas that persist over time are almost exclusively structural and seizing-related, reinforcing that the choke point is systematic rather than episodic. The persistence of dilemmas like D02 (Bureaucratic Process vs. Urgent Agile Execution) suggests that these are not problems to be 'solved' once, but enduring tensions that must be continuously managed.

5. Discussion

Our findings provide a micro-level, process-based view of project governance in DT initiatives, with significant implications for both theory and practice. We found that governance tensions are not random but concentrate at the intersection of structural constraints and seizing-related decisions, creating a predictable choke point. In this section, we discuss the theoretical contributions of these findings and their practical implications for managing DT projects.

5.1. Theoretical Contributions

This study makes three main contributions to the literature on project management, digital transformation, and dynamic capabilities.

First, we empirically specify the micro-foundations of the 'structural dilemma' in DT projects. Although the existing literature generally suggests that agile project management can play a significant role in digital transformation by enhancing organizational flexibility, fostering cross-functional team collaboration, and supporting continuous learning and iterative innovation [5,10,13,14,17], and has also identified the macro-level tension between agility and stability as a key challenge [27,38], our study unpacks how this manifests in the mundane, day-to-day reality of project meetings. Dilemmas like D02 (Bureaucracy vs. Agility), D09 (Control vs. Autonomy), and D10 (Standardization vs. Customization) are the concrete, observable micro-practices through which the macro-structural dilemma is enacted and navigated. Our finding that 62% of all identified dilemmas map to this macro-category provides strong evidence that organizational structure is the primary battleground where DT governance is contested.

Second, we refine the understanding of dynamic capabilities in the context of corporate innovation. While Teece's [59] framework gives equal weight to sensing, seizing, and reconfiguring, our findings reveal a significant concentration of governance dilemmas at the seizing stage (69%). This suggests that for corporate ventures, which often leverage the parent's existing sensing capabilities, the most critical bottleneck is not opportunity recognition but decision-making and resource commitment. This 'seizing choke point' highlights the social and political challenges of mobilizing resources and securing buy-in within a complex parent organization, a dimension often under-theorized in the dynamic capabilities literature [67,79].

Third, we propose two meso-level governance dilemmas that mediate between macro-structures and micro-interactions (see Figure 4). Our analysis showed that many micro-dilemmas were cross-cutting, mapping to multiple macro-dilemmas. This led us to synthesize two meso-level dilemmas particularly salient in the incubation context: (a) the Stakeholder Engagement Dilemma—the tension between deep engagement with specific early-adopter stakeholders and the need to build a scalable, standardized solution for a broader market [80]; and (b) the Compliance vs. Agility Dilemma—the tension between adhering to the parent company's formal processes and the need for agile, experimental methods required for rapid innovation [81]. These meso-level dilemmas act as transmission mechanisms, translating abstract macro-structural pressures into the concrete micro-level decision tensions faced by project teams.

5.2. Practical Implications: An AI-Enabled Early-Warning System

Beyond its theoretical contributions, this study offers a replicable, AI-enabled methodology that can be operationalized as an early-warning system for project governance tensions. Project and portfolio managers can use a similar meeting-analytics pipeline to monitor the health of their DT projects in near real-time. By automatically analyzing the transcripts of governance meetings, the system can: (1) identify emerging dilemmas before they escalate into major roadblocks; (2) quantify governance tension profile of a project, highlighting whether its primary challenges are related to sensing, seizing, or reconfiguring; and (3) benchmark across projects to identify systemic, portfolio-level governance issues that require intervention from senior leadership.

For example, a project that consistently flags the Bureaucratic Process vs. Urgent Agile Execution dilemma (D02) may need a dedicated process concierge or a pre-approved budget for expedited procurement. A portfolio with a high concentration of seizing-related dilemmas may indicate a need to clarify decision rights or streamline the resource allocation process at the corporate level. By making invisible governance tensions visible and quantifiable, the AI-enabled pipeline empowers managers to move from reactive problem-solving to proactive governance design, thereby reducing decision latency and improving project delivery performance [32].

6. Conclusions

This study set out to investigate the project governance dilemmas that emerge in the day-to-day interactions of corporate DT initiatives. Through a micro-longitudinal, AI-assisted analysis of 28 governance meetings across two anonymized corporate startup programs (Startup 1 and Startup 2), we identified 13 distinct dilemmas, revealing a critical governance choke point at the intersection of structural constraints and seizing-related decisions. Our findings contribute to a more dynamic, process-oriented, and data-driven understanding of project governance, with clear implications for both theory and practice.

Like all studies, this one has limitations. Our analysis was based on two projects within a single technology firm, which may limit the generalizability of the specific dilemmas identified. Future research should apply this methodology across a broader range of industries and organizational contexts. Furthermore, our analysis is based on what was said in formal governance meetings; it does not capture the informal, offline conversations and political maneuvering that also shape project outcomes. Tensions outside meetings and implicit tensions may be underrepresented, a limitation that future work could address by estimating false-negative rates through systematic sampling of negative episodes. Finally, while our LLM-based approach proved effective, it is a conservative method that may under-represent the full extent of subtly expressed tensions.

Despite these limitations, this study provides a novel and powerful approach to understanding and managing the complex governance challenges of digital transformation. It demonstrates that the path to successful DT is paved not by eliminating tensions, but by building the organizational capacity to navigate them effectively. The AI-enabled tools and conceptual frameworks developed here offer a promising step in that direction, enabling managers and researchers alike to better understand the micro-foundations of project performance in the digital age.