From AI Adoption to Sustainable Value

Abstract

Artificial intelligence (AI) has become an increasingly important component of digital transformation, with organisations adopting AI technologies to improve productivity, automate processes, enhance decision-making, and create new sources of competitive advantage. However, the adoption of AI does not automatically translate into sustainable enterprise value. This study examines how organisations can move from AI adoption to sustainable value creation through a meta-study of recent (2021-2026) peer-reviewed literature. Rather than collecting primary empirical data, the study integrates and critically synthesises findings from recent research on AI adoption, organisational capabilities, business process integration, human–AI collaboration, AI governance, and organisational performance.

The literature synthesis indicates that successful AI value realisation is not primarily a technological challenge but an organisational and strategic one. Six interrelated capabilities emerge as particularly important: strategic alignment, operating-model capability, data and systems integration, human and organisational capability, responsible AI governance, and value-management capability. The literature further suggests that organisations frequently achieve initial efficiency and productivity improvements before realising broader strategic benefits. The transition from AI experimentation and pilot projects to scalable, enterprise-wide value therefore depends on embedding AI within business processes, organisational structures, decision-making systems, and governance arrangements. Evidence also indicates that employee skills, organisational readiness, and effective human–AI collaboration are critical to determining whether AI capabilities translate into improved organisational outcomes.

The study proposes an integrated perspective of AI value realisation in which sustainable enterprise value emerges from the interaction of technological capabilities with organisational, human, process, strategic, and governance capabilities. The findings contribute to the growing literature on AI and business value by shifting attention from AI adoption as an end in itself towards the organisational conditions required to convert AI investments into sustained performance and strategic value. For managers, the study highlights the importance of treating AI as an organisational transformation capability rather than simply as a technology investment.

Keywords: artificial intelligence; AI adoption; enterprise value; organisational capabilities; digital transformation; human–AI collaboration; AI governance; business process integration; sustainable value creation.

1. Introduction

1.1 Background and context

Artificial intelligence (AI) has become an important component of contemporary business strategy. Organizations increasingly use AI to automate tasks, support decision-making, improve customer interactions and enhance operational efficiency. Yet the adoption of AI does not, in itself, guarantee improved organizational performance. The more difficult challenge is to embed AI into the structures, processes and capabilities through which firms create and capture value.

This distinction is central to The AI Trends Report: How to turn AI into business value in 2026 and beyond (OMMAX, Ibexa and Make, 2026). Based on a survey of 250 decision-makers across France, Germany, Italy, the Netherlands and the United Kingdom, the report examines AI maturity, operating models, business impact, scaling barriers, data foundations, governance and organizational enablement. It finds that 58% of organizations have a fully defined AI strategy, but only 44% have a fully implemented AI operating model. The report therefore identifies a gap between strategic ambition and the organizational arrangements required to execute it.

This gap provides the starting point for the present study. The central issue is not whether organizations recognize the importance of AI, but whether they possess the capabilities necessary to translate that recognition into measurable and sustainable outcomes. The report’s findings suggest that AI transformation is increasingly an organizational challenge rather than a purely technological one. Its emphasis on operating models, integration, governance and business ownership indicates that the realization of AI value depends on how technology is connected to the wider enterprise.

1.2 The problem of AI value creation

The literature on AI capability provides a useful foundation for understanding this challenge. Mikalef and Gupta (2021) conceptualize AI capability as the combination of technological, human and organizational resources that enables firms to use AI effectively. Their empirical study finds that AI capability is associated with organizational creativity and firm performance, suggesting that the benefits of AI arise from complementary capabilities rather than from technological adoption alone.

This perspective is particularly relevant to the report’s finding that AI strategy is more advanced than AI operating-model implementation. A strategy may establish priorities and ambitions, but an operating model determines how responsibilities, processes, resources and governance are organized to deliver them. The report’s recommendation to align operating models with strategy therefore reflects a broader theoretical argument: AI creates value when organizations develop the capacity to integrate it into their existing systems of work and decision-making.

The problem is also evident in the transition from experimentation to production. The report finds that 44% of AI failures occur between pilot and production, while 35% occur during the pilot itself. It also reports that 65% of AI projects exceed their initial budgets. These findings suggest that the difficulty of AI transformation lies not simply in developing a promising application, but in sustaining it within the o.perational and financial realities of the organization.

Aydiner et al. (2022) provide a complementary explanation. Their study identifies AI project planning, co-development, data management and AI model lifecycle management as organizational capabilities required to address the uncertainty and data dependency of AI implementation. Their findings reinforce the argument that successful AI deployment requires more than technical development; it requires organizational arrangements capable of managing the distinctive challenges of AI throughout its lifecycle.

1.3 Efficiency gains and the limits of adoption

The report identifies operational efficiency as the strongest area of AI impact. It finds that 84% of respondents report some or strong increases in operational efficiency, compared with 75% reporting revenue increases. It also reports that 55% of organizations have achieved efficiency improvements of at least 10%, while 36% report revenue improvements above 10%. These findings suggest that AI is currently delivering its most visible returns through the optimization of internal activities.

Peer-reviewed research supports the importance of this efficiency effect, while also demonstrating that its magnitude depends on the context of use. Brynjolfsson, Li and Raymond (2025), in a study of 5,172 customer-support agents, find that generative AI assistance increases issues resolved per hour by 15%, with larger gains among less experienced workers. Their findings suggest that AI can improve productivity while also supporting the diffusion of knowledge across an organization.

However, the relationship between AI and productivity is not uniform. Dell’Acqua et al. (2025) describe a “jagged technological frontier” in which AI improves performance on some tasks but can reduce performance on others. Their study of knowledge workers demonstrates that AI users can complete tasks faster and produce higher-quality solutions when the tasks fall within the technology’s capabilities. This finding is important because it cautions against treating AI adoption as a reliable proxy for organizational improvement.

The report’s efficiency-first finding should therefore be interpreted as a description of current business outcomes rather than as evidence that AI inevitably improves performance. The strategic challenge is to identify where AI complements human work, where it requires additional oversight, and where its benefits justify the costs of implementation. The report itself argues that efficiency gains should provide a foundation for broader value creation, including commercial and customer-facing applications.

1.4 From pilot projects to production-scale transformation

A central concern of this study is the difficulty of moving AI from isolated experimentation to production-scale deployment. The report’s finding that the largest share of failures occurs between pilot and production suggests that organizations face a transition problem: the conditions that support experimentation are not necessarily the conditions required for operational sustainability.

This distinction is consistent with the AI implementation literature. Aydiner et al. (2022) argue that AI projects require capabilities for planning, co-development, data management and lifecycle management because AI systems are probabilistic, data-dependent and subject to change. These characteristics create implementation challenges that differ from those associated with more conventional information systems.

The report also identifies integration complexity as the leading barrier to stronger AI impact, cited by 40% of respondents. It argues that scaling AI requires the coordination of systems, workflows, teams and decision rights, rather than simply improving the technology stack.

This suggests that AI transformation is fundamentally a process of organizational integration.

Mikalef et al. (2021) provide empirical support for this interpretation. Their study of 448 EU organizations finds that AI adoption improves organizational performance through decision-making and business process performance. Process automation, organizational learning and process innovation act as complementary mechanisms through which AI creates value. The implication is that AI becomes economically meaningful when it improves the processes through which organizations operate, learn and make decisions.

1.5 Data integration, governance and organizational readiness

The report’s discussion of data and infrastructure further demonstrates that AI value depends on complementary organizational capabilities. It finds that data quality is the leading reason AI initiatives fail to scale, cited by 28% of respondents, while integration complexity remains a major obstacle. It also reports that data warehouses and lakes are less fully integrated for AI-driven workflows than several front-office systems.

These findings reinforce the argument that AI cannot be separated from the quality and accessibility of the data on which it depends. Aydiner et al. (2022) identify data management as a core AI implementation capability, emphasizing that data must be governed and maintained throughout the AI lifecycle. From this perspective, data integration is not merely a technical prerequisite; it is part of the organizational infrastructure through which AI value is created.

Governance is equally important. The report finds that 66% of organizations assess AI risks often or very often, but only 46% report fully implemented governance for agentic AI. It argues that autonomous and sequential AI systems require real-time monitoring, audit trails and escalation protocols that extend beyond traditional deployment controls.

This concern is consistent with the emerging literature on responsible AI governance. Embedding AI Governance in Organizations argues that governance requires the translation of external ethical and regulatory principles into internal organizational practices, supported by strategic direction, operational execution and coordination across functions. The report’s emphasis on governance maturity therefore reflects a wider scholarly concern: AI value must be created within structures that make its use accountable, controllable and sustainable.

Organizational readiness also depends on who owns AI. The report finds that 48% of AI execution sits within IT and engineering, compared with only 7% in business units. It suggests that this concentration may help explain why efficiency gains are more visible than commercial outcomes.

This observation is consistent with the argument that AI transformation requires both technical expertise and business accountability. If AI remains primarily a technical initiative, organizations may optimize processes without adequately redefining customer value, product strategy or revenue generation.

1.6 Research aim and objectives

The aim of this study is to examine how organizations can convert AI adoption into sustainable enterprise value by developing the organizational capabilities required for effective implementation and scaling.

The study has four objectives. First, it seeks to examine the relationship between AI strategy and the organizational operating models required to execute it. Second, it aims to investigate the role of data integration and business process performance in translating AI adoption into organizational outcomes. Third, it seeks to explore how governance and organizational readiness influence the responsible scaling of AI. Finally, it aims to assess how AI efficiency gains can be converted into broader commercial and enterprise value.

These objectives are informed by the report’s central finding that the challenge is no longer simply access to AI, but execution across systems, data and organizational structures.

1.7 Research question

The study is guided by the following research question:

How can organizations develop the capabilities required to convert AI adoption into sustainable enterprise value?

This question is deliberately broader than the measurement of AI adoption. It focuses on the organizational conditions that enable AI to move beyond experimentation and contribute to measurable performance, responsible governance and long-term value creation.

1.8 Significance of the study

The study is significant because it addresses a central challenge in contemporary AI research and practice: the gap between technological potential and organizational realization. The report provides a timely empirical account of AI maturity among European organizations, while the peer-reviewed literature offers theoretical and empirical foundations for understanding why AI value depends on complementary capabilities.

The study contributes to this discussion by bringing together the report’s findings on operating models, efficiency, scaling, integration, governance and ownership. It also recognizes that AI productivity gains are context-dependent and that the transition from pilot to production involves organizational, technical and financial challenges. By examining these issues together, the study seeks to provide a more comprehensive understanding of AI as a capability for enterprise transformation rather than merely a technology for automation.

1.9 Scope and limitations

The study focuses on AI adoption and value creation within organizational settings. It considers AI strategy, operating models, business processes, data integration, governance and organizational readiness as interrelated factors influencing enterprise value. It does not attempt to evaluate individual AI models or compare the technical performance of specific AI systems.

The report provides the contemporary empirical context for the study, but its survey findings should be interpreted as descriptive rather than causal. The sample consists of 250 decision-makers across five European countries, and the report does not establish that particular organizational practices directly cause better AI outcomes. The peer-reviewed literature is therefore used to develop and substantiate the study’s theoretical argument, while the report is used to illustrate the practical relevance of that argument.

1.10 Chapter structure

The remainder of the study is organized as follows. Chapter 2 reviews the literature on AI capability, organizational transformation, business process performance and responsible AI governance. Chapter 3 explains the research methodology and the approach used to examine the research question. Chapter 4 presents the findings, while Chapter 5 discusses those findings in relation to the literature and the report’s main themes. Chapter 6 concludes the study by outlining its implications, limitations and directions for future research.

2. Literature Review

2.1 Introduction

The purpose of this chapter is to establish the theoretical and empirical foundations for examining how organisations convert artificial intelligence (AI) adoption into measurable business value. Whereas Chapter 1 introduced the research problem and identified the gap between AI ambition and organisational execution, this chapter reviews the literature that explains why this gap emerges and how it may be addressed.

The literature increasingly suggests that AI should not be understood as a standalone technological investment. Rather, its value depends on the interaction between technological resources, organisational capabilities, business processes, data foundations, governance arrangements and employee participation. This perspective is particularly important because organisations may successfully deploy AI applications without necessarily achieving sustained improvements in productivity, revenue or competitive performance. The central concern is therefore not simply whether AI is adopted, but whether it is embedded effectively into the organisational system.

The chapter first defines AI business value and distinguishes between adoption, implementation and value realisation. It then reviews the role of organisational capabilities, strategic alignment, data and systems integration, governance, organisational readiness and human–AI collaboration. The chapter concludes by identifying the principal gaps in the existing literature and explaining how these gaps inform the present research.

2.2 Artificial Intelligence and Business Value

Artificial intelligence is a broad field encompassing technologies that enable machines to perform activities associated with human intelligence, including perception, learning, reasoning, prediction, decision-making and language processing. In organisational contexts, AI may include machine learning, natural language processing, computer vision, recommender systems, intelligent automation and generative AI. These technologies differ in their technical characteristics and applications, but they share the potential to augment or automate activities that were previously dependent on human judgement or manual execution.

The relationship between AI and business value is not automatic. Enholm et al. (2022) argue that AI business value must be understood through the mechanisms by which AI is adopted, used and embedded in organisational activities. Their systematic literature review distinguishes between the enablers and inhibitors of AI use, the types of organisational applications and the first- and second-order effects through which value is created. First-order effects occur at the process level, such as faster task completion, improved forecasting or reduced error rates. Second-order effects occur at the organisational level, such as improved profitability, innovation, customer experience or competitive advantage.

This distinction is important because the immediate effects of AI may not be equivalent to its broader business outcomes. For example, an AI system may reduce the time required to complete a task, but the organisation will only obtain a meaningful business benefit if the released capacity is redirected towards productive activity, improved service or strategic growth. Similarly, a model may produce accurate predictions without improving performance if employees do not trust the output, if the recommendation is not integrated into workflows or if decision rights remain unclear.

AI business value can therefore be understood as the outcome of a transformation process rather than as a direct consequence of technology adoption. The process involves identifying a valuable use case, developing or acquiring an appropriate AI solution, integrating it with organisational systems, redesigning work practices, enabling employees to use it effectively and monitoring whether the expected outcomes are achieved. This interpretation is consistent with the literature’s increasing emphasis on complementary resources and organisational capabilities.

The distinction between technological potential and realised value is also reflected in the 2026 AI Trends Report. The report finds that AI is already associated with operational efficiency and revenue impact, but that efficiency gains are more widespread than revenue gains. Among organisations with AI use cases in production or at scale, 55% report efficiency improvements of at least 10%, compared with 36% reporting revenue improvements above 10%. The report consequently presents efficiency as an early and measurable return, while positioning revenue and growth as a subsequent stage of maturity (OMMAX, Ibexa and Make, 2026).

2.3 AI Adoption, Implementation and Value Realisation

A central issue in the literature is the need to distinguish between AI adoption and AI implementation. Adoption generally refers to the decision to accept or acquire a technology, whereas implementation concerns the practical process of embedding it into organisational activities. Value realisation goes further by examining whether implementation produces the intended operational, financial or strategic outcomes.

This distinction is necessary because an organisation may adopt AI without integrating it into its core processes. A pilot project, proof of concept or isolated departmental application may demonstrate technical feasibility but still fail to generate enterprise-level value. Lee et al. (2023), in their systematic literature review of AI implementation, identify organisational, information systems, technological and people-related dimensions as important influences on implementation outcomes. Their findings indicate that AI implementation is a multidimensional organisational process rather than a purely technical exercise.

The implementation challenge is particularly significant because AI systems are often dependent on data availability, data quality, system interoperability and ongoing human oversight. Unlike some conventional software applications, AI systems may also require continuous monitoring, retraining and adjustment as data, customer behaviour and organisational conditions change. Implementation is therefore not a one-time event but an ongoing lifecycle involving development, deployment, use, evaluation and improvement.

The 2026 AI Trends Report provides evidence of this implementation gap. Although 58% of surveyed organisations report having a fully defined AI strategy, only 44% report having a fully implemented end-to-end AI operating model. A further 52% describe their operating model as partly implemented. The report interprets this difference as evidence that strategic ambition is developing faster than the organisational mechanisms required to deliver it (OMMAX, Ibexa and Make, 2026).

The report also indicates that 79% of AI initiatives fail during or after the pilot stage, while 65% exceed their budgets. These findings suggest that technical experimentation is not the primary determinant of success. The more difficult challenge lies in moving from experimentation to reliable production use and then from individual use cases to scalable organisational deployment.

Consequently, AI maturity should not be assessed solely through the number of pilots, models or applications developed. A more meaningful assessment must consider whether AI is connected to business priorities, supported by appropriate operating structures, integrated into existing systems, governed throughout its lifecycle and used consistently by employees and decision-makers.

2.4 The Organisational Capability Perspective

The organisational capability perspective provides a useful theoretical foundation for explaining why some organisations realise more value from AI than others. From this perspective, resources such as algorithms, computing infrastructure, data and technical expertise are necessary but insufficient. Value depends on the organisation’s ability to combine and deploy these resources effectively.

Mikalef and Gupta (2021) conceptualise AI capability as an organisation’s ability to mobilise and orchestrate AI-related resources in ways that support organisational creativity and performance. Their approach moves beyond the possession of technical assets and emphasises the importance of tangible resources, human skills and organisational resources. Tangible resources include data, technological infrastructure and AI tools. Human resources include technical knowledge, analytical ability and domain expertise. Organisational resources include structures, processes, coordination mechanisms and managerial support.

This perspective is consistent with the knowledge-based view of the firm, which argues that competitive advantage depends not simply on possessing resources but on integrating and applying knowledge through organisational routines. In the AI context, the relevant capability is therefore not merely the ability to build a model. It is the ability to identify appropriate opportunities, prepare data, develop or select a solution, integrate it into workflows, manage risks and convert its outputs into decisions or actions.

Aydiner et al. (2022) similarly argue that AI-specific resources do not create value independently. Instead, organisations must develop capabilities that combine human, technological and intangible resources. Their analysis highlights the importance of coping with two distinctive characteristics of AI: inscrutability and data dependency. Inscrutability refers to the difficulty of understanding or explaining how some AI systems produce their outputs. Data dependency refers to the extent to which AI performance relies on the availability, quality, relevance and timeliness of data.

These characteristics create organisational requirements that are less pronounced in many traditional information systems. Organisations need capabilities for interpreting model outputs, validating results, managing exceptions, ensuring data quality and assigning responsibility for AI-supported decisions. AI capability therefore includes both technical competence and the organisational capacity to use AI responsibly and productively.

The capability perspective also helps explain why similar AI technologies may produce different outcomes across organisations. Two organisations may acquire comparable tools, but the organisation with stronger data management, clearer decision rights, better process integration and more developed employee skills is more likely to generate sustained value. The difference lies not in the technology itself, but in the complementary capabilities surrounding it.

2.5 Strategic Alignment and the AI Operating Model

Strategic alignment concerns the relationship between technology initiatives and organisational objectives. In the AI context, alignment requires organisations to identify where AI can contribute to strategic priorities and to establish mechanisms for selecting, prioritising and evaluating use cases.

AI initiatives may fail when they are selected because a technology is fashionable rather than because a business problem has been clearly defined. A technically impressive application may have limited value if it addresses a low-priority activity, lacks an identifiable owner or cannot be connected to measurable outcomes. Conversely, relatively modest applications may produce significant value when they address high-volume processes, recurring customer needs or important operational bottlenecks.

The literature therefore emphasises the importance of business-led use-case identification and cross-functional collaboration. AI projects often require knowledge from several domains, including IT, data, operations, legal, risk, marketing and frontline functions. Strategic alignment cannot be achieved by technology departments alone because the value of AI depends on how its outputs affect business processes and decisions.

The 2026 AI Trends Report identifies a similar imbalance between strategic intent and operational execution. While a majority of respondents report a defined AI strategy, ownership remains concentrated in IT and engineering. IT or engineering functions hold primary AI ownership in 48% of organisations, whereas only 7% report that ownership sits primarily within business units. The report interprets this pattern as evidence that AI remains predominantly technology-led, despite the need for stronger business ownership and accountability.

An effective AI operating model should clarify at least five issues. First, it should establish who identifies and prioritises use cases. Second, it should define who owns the business outcome. Third, it should allocate responsibility for data, model development and deployment. Fourth, it should specify governance and escalation procedures. Fifth, it should establish how performance and return on investment will be monitored.

The operating model is therefore the organisational mechanism through which strategy becomes execution. Without it, AI initiatives may remain fragmented, duplicated or disconnected from enterprise priorities. A defined strategy provides direction, but an operating model provides the roles, processes and coordination mechanisms required to deliver that strategy.

2.6 Data Quality, Integration and Infrastructure

Data is a foundational requirement for AI because models depend on data to learn, generate predictions and produce recommendations. However, the availability of data does not guarantee its usefulness. Data must also be accurate, complete, timely, relevant, accessible and appropriately governed.

Enholm et al. (2022) identify data quality, data availability and data infrastructure as important technological enablers of AI use. Poor-quality data can reduce model accuracy, create inconsistent outputs and undermine user trust. Fragmented data can also prevent AI systems from developing a sufficiently complete view of customers, products, operations or business performance.

Data integration is particularly important when AI is intended to support cross-functional workflows. A model may operate effectively in isolation but fail to create value if it cannot access the systems required to execute its recommendations. For example, a customer insight may be generated by an AI system but remain commercially irrelevant if it is not connected to the customer relationship management system, marketing platform or customer service process.

The 2026 AI Trends Report identifies integration complexity as the leading barrier to AI scaling, cited by 40% of respondents. Data quality is identified as the number-one reason AI initiatives fail to scale by 28% of respondents. The report also finds that front-office systems are more likely to be fully integrated than data lakes and warehouses, which remain a bottleneck for AI-driven workflows.

These findings support the argument that AI scalability depends on the wider data and systems architecture. Organisations may be able to launch isolated pilots using manually prepared datasets, but enterprise-scale deployment requires reliable data pipelines, interoperable systems and repeatable integration practices. The challenge is not simply to improve individual datasets, but to establish a data foundation that supports multiple use cases across functions.

This also explains why integration should often precede automation. Automating a fragmented process may increase speed without improving the overall outcome. In some cases, it may reproduce existing errors more quickly or create new coordination problems. A more sustainable sequence is to integrate relevant data and systems, automate clearly defined activities and then orchestrate more complex cross-functional workflows.

2.7 Governance, Risk and Responsible AI

AI governance refers to the structures, policies, processes and controls through which organisations direct, monitor and regulate the development and use of AI. Governance is relevant not only to legal compliance but also to performance, trust, accountability and risk management.

AI systems can introduce risks involving privacy, discrimination, security, explainability, intellectual property, operational resilience and inappropriate decision-making. These risks may arise at different stages of the AI lifecycle, including data collection, model development, deployment, monitoring and retirement. Governance must therefore extend beyond approval at the beginning of a project and continue throughout the system’s use.

Enholm et al. (2022) argue that responsible AI governance should be embedded across the design, deployment and evaluation of AI applications. Governance should not be treated as an external constraint imposed after technical development. Instead, it should be integrated into the way AI initiatives are selected, designed and managed.

The organisational nature of AI governance is also important. Responsibility cannot rest solely with technical teams because many AI risks are connected to business context, customer impact and organisational decision-making. Effective governance requires collaboration between technical specialists, business owners, legal and compliance functions, risk professionals and affected employees.

The 2026 AI Trends Report indicates that AI governance is widely established but not yet fully mature. Sixty-six percent of respondents report assessing AI risks often or very often, while only 46% report that governance for agentic AI is fully implemented. The report also finds that 80% of organisations share sensitive data externally with AI providers, while 54% use flexible, use-case-based data sovereignty policies.

These findings suggest a tension between the need for rapid AI adoption and the need for consistent control. Flexible governance may allow organisations to experiment and respond to different use cases, but it may also create ambiguity if data classification, provider responsibilities and approval requirements are not clearly defined. Governance must therefore be sufficiently flexible to accommodate different applications while remaining sufficiently consistent to protect the organisation and its stakeholders.

2.8 Organisational Readiness and Employee Enablement

Organisational readiness refers to the extent to which an organisation possesses the resources, structures, skills and willingness required to implement and sustain a new technology. In AI implementation, readiness includes technological preparedness, managerial support, employee capabilities, organisational culture and the capacity to redesign work.

AI can alter tasks, responsibilities and decision-making processes. Employees may need to interpret model outputs, supervise automated activities, challenge recommendations or develop new forms of domain expertise. As a result, AI implementation requires more than technical training. It also requires communication, participation, trust-building and clarity about how work will change.

The literature suggests that employee acceptance is influenced by whether AI is perceived as useful, understandable and compatible with existing work practices. Employees may resist systems that appear to threaten their autonomy, increase monitoring or introduce accountability without sufficient control. Conversely, employees may support AI when it reduces repetitive work, improves decision quality and allows them to focus on more valuable activities.

Organisational learning is therefore a critical component of AI capability. Organisations must develop mechanisms for sharing knowledge between teams, documenting lessons from pilots, identifying reusable components and disseminating successful practices. Without these mechanisms, AI learning remains localised and the organisation repeatedly solves similar problems in different departments.

The 2026 AI Trends Report identifies several areas in which enablement remains incomplete. Full implementation is reported for return-on-investment tracking by 53% of respondents, AI champion programmes by 50%, knowledge-sharing mechanisms by 46% and employee enablement by 39%.

These results indicate that organisational enablement is often treated as a secondary activity rather than as part of the infrastructure of AI deployment. However, if employees lack the skills or confidence to use AI, technical deployment may not translate into operational adoption. Similarly, if knowledge is not shared across teams, organisations may fail to scale successful use cases or may duplicate unsuccessful experiments.

2.9 Human–AI Collaboration and the Limits of Automation

The relationship between AI and human work is more complex than a simple substitution of machines for people. AI may automate certain tasks, augment human judgement or create new forms of collaboration. The outcome depends on the nature of the task, the quality of the AI system, the expertise of the employee and the design of the surrounding workflow.

Research on generative AI illustrates this complexity. Brynjolfsson, Li and Raymond (2025) find that generative AI can improve worker productivity, particularly for less experienced employees, but the effects vary across workers and tasks. Dell’Acqua et al. (2025) similarly describe a “jagged technological frontier” in which AI performs strongly on some tasks but poorly on others. Their findings suggest that productivity gains depend on understanding where AI is reliable and where human expertise remains essential.

This has important implications for organisational design. AI should not necessarily be introduced with the assumption that every task can or should be automated. Organisations must determine which activities are suitable for automation, which require human review and which are best supported through decision augmentation. In high-risk or ambiguous contexts, human oversight may be necessary even when the AI system appears technically capable.

The literature therefore supports a human-centred approach to AI implementation. The objective is not simply to maximise automation, but to design effective combinations of human judgement and machine capability. This may involve human-in-the-loop processes, escalation mechanisms, exception handling, employee training and continuous evaluation of system performance.

2.10 From Efficiency Gains to Revenue and Strategic Value

The literature commonly distinguishes between operational and strategic forms of AI value. Operational value includes cost reduction, productivity improvement, faster processing, fewer errors and improved resource utilisation. Strategic value includes revenue growth, innovation, customer experience, resilience, market differentiation and new business models.

Operational benefits are often easier to measure because they can be linked to time, cost or throughput. Strategic benefits are more difficult to isolate because they may emerge gradually and depend on multiple organisational and market conditions. This difference may explain why organisations frequently report efficiency gains before they report substantial revenue impact.

The 2026 AI Trends Report reflects this pattern. Productivity is identified as the leading AI value driver, and IT and software development are reported as the functions generating the highest realised returns. At the same time, the report argues that efficiency gains should be treated as a foundation for further value creation rather than as the final objective. Freed capacity must be redirected towards customer-facing innovation, commercial activity and strategic differentiation if AI is to contribute to sustained growth.

This argument is consistent with the broader literature on complementary organisational resources. Productivity improvements may create the conditions for growth, but they do not automatically produce growth. Revenue impact requires organisations to redesign customer journeys, improve personalisation, develop new offerings or integrate AI into commercial decision-making. Consequently, the transition from efficiency to revenue requires a second stage of organisational transformation.

2.11 Synthesis of the Literature

The literature reviewed in this chapter identifies several interdependent conditions for AI value realisation.

First, AI value depends on organisational capabilities rather than on technological assets alone. Data, algorithms and infrastructure must be combined with human expertise, managerial support and organisational routines.

Second, strategic alignment is essential. AI initiatives are more likely to create value when they address clearly defined business priorities and have identifiable owners responsible for outcomes.

Third, implementation requires integration. AI systems must be connected to relevant data sources, business applications and operational workflows. Fragmented systems and poor data quality can prevent technically successful applications from producing meaningful results.

Fourth, governance must be embedded throughout the AI lifecycle. Responsible AI requires clear accountability, risk assessment, data controls, monitoring and mechanisms for human oversight.

Fifth, organisational readiness and employee enablement are necessary for sustained adoption. Employees need skills, confidence, participation and clarity about how AI changes their work.

Finally, AI value develops through stages. Initial benefits often appear as operational efficiency, while broader revenue and strategic benefits require deeper process redesign, cross-functional integration and organisational learning.

These themes indicate that the AI value problem is fundamentally socio-technical. It cannot be explained through technology alone, nor through organisational factors in isolation. AI value emerges from the interaction between technology, data, people, processes, governance and strategy.

2.12 Research Gap

Although the recent peer-reviewed literature provides an increasingly comprehensive understanding of artificial intelligence (AI) adoption, implementation and organisational value, several gaps remain in the way these findings are connected and synthesised.

First, the literature provides substantial evidence on the capabilities and organisational conditions associated with AI adoption, but these factors are frequently examined independently. Less attention has been given to how strategic alignment, operating-model capability, data and systems integration, human and organisational capability, governance, and value-management mechanisms interact as organisations progress from AI experimentation and pilot initiatives towards scaled enterprise implementation.

Second, existing research remains fragmented across different levels of analysis. Studies variously focus on technological capabilities, organisational readiness, business processes, employee skills, leadership, governance or organisational performance. While these perspectives provide valuable insights, there is a need for a more integrated explanation of how these complementary capabilities collectively contribute to AI value realisation. In particular, the literature would benefit from a synthesis that connects technology adoption with the organisational capabilities required to embed AI into routine business operations.

Third, the literature indicates that the benefits of AI are not necessarily realised at the same level or at the same stage of adoption. Productivity and operational efficiency are often among the most immediate outcomes, whereas broader strategic benefits, such as innovation, customer value and competitive advantage, require deeper organisational integration. However, the relationship between these different forms of value remains insufficiently integrated within the existing literature. A clearer understanding is therefore required of how organisations can progress from initial efficiency improvements towards broader and more sustainable forms of enterprise value.

Fourth, the growing literature on responsible AI highlights the importance of governance, accountability, risk management and appropriate organisational controls. However, governance is often considered separately from AI implementation and value creation. This creates a gap in understanding how responsible AI governance can operate not only as a mechanism for controlling risk, but also as an organisational capability that supports the responsible scaling and sustainable use of AI.

Finally, despite the rapid expansion of research on AI and organisational performance, the literature remains heterogeneous in its conceptualisations of AI capability, adoption, implementation success and business value. Recent studies use different theoretical perspectives, measures and organisational contexts, making it difficult to derive a coherent explanation of the conditions under which AI produces sustainable value. There is therefore a need for an integrative synthesis of recent peer-reviewed research that brings these perspectives together and identifies the common organisational mechanisms underlying successful AI value realisation.

The present study addresses this gap through a meta-study of recent peer-reviewed literature. Rather than treating AI adoption as an isolated technological decision, the study examines AI value realisation as an organisational and strategic process. It synthesises evidence concerning strategic alignment, operating-model capability, data and systems integration, human and organisational capability, responsible AI governance and value management, with the aim of developing an integrated understanding of how organisations can move from AI adoption and experimentation towards scalable and sustainable enterprise value.

2.13 Chapter Summary

This chapter reviewed the recent peer-reviewed literature on AI adoption, organisational capabilities, implementation, business processes, human–AI collaboration, governance and business value. The literature indicates that AI adoption alone does not guarantee improved organisational performance. Instead, value realisation depends on the development of complementary organisational capabilities that enable AI technologies to be integrated into business processes, decision-making structures and organisational routines.

The review identified several recurring themes across the literature. Strategic alignment and business ownership influence whether AI initiatives address meaningful organisational priorities. Data quality, systems integration and business-process capabilities provide important foundations for implementation and scaling. Organisational readiness, employee skills and human–AI collaboration influence how effectively AI is embedded in everyday work. Responsible governance provides mechanisms for managing risk, accountability and appropriate use while supporting sustainable adoption.

The literature also suggests that AI value develops progressively. Initial benefits are frequently associated with productivity, automation and operational efficiency, while broader strategic value requires deeper integration with organisational processes, capabilities and customer-facing activities. This indicates that the transition from AI adoption to sustainable enterprise value is not simply a technological progression, but an organisational transformation process.

However, the literature remains fragmented, with different studies examining individual technological, organisational, human, process and governance factors in isolation. The central research gap therefore concerns the absence of an integrated synthesis explaining how these complementary capabilities interact across the journey from AI adoption and experimentation to scalable and sustainable value creation.

Accordingly, the subsequent chapters build on this gap by conducting a meta-study of recent peer-reviewed literature. The analysis focuses on the relationships among strategic alignment, operating-model capability, data and systems integration, human and organisational capability, responsible AI governance and value management. This provides the foundation for developing an integrated perspective on AI value realisation and for explaining why organisations may achieve successful AI adoption without necessarily achieving sustainable enterprise value.

3. Research Methodology

3.1 Introduction

This chapter sets out the methodology used to examine how organisations convert artificial intelligence (AI) adoption into sustainable enterprise value through a meta-study of recent peer-reviewed literature. The methodological shift is deliberate. The research question is concerned with an organisational phenomenon that has already generated a rapidly expanding body of scholarly work across information systems, management, operations, organisational behaviour and technology-management journals. Rather than collecting primary data from organisations, the present study synthesises this dispersed evidence to identify recurring explanations, areas of convergence, boundary conditions and unresolved questions.

A literature-based design is appropriate because AI value realisation is not adequately explained by adoption rates alone. Recent research examines AI capability, organisational readiness, implementation, business-process performance, decision-making, human–AI collaboration, governance and scaling as interdependent issues. Systematic reviews have similarly demonstrated that AI implementation is a multidimensional organisational process involving technological, organisational, information-system and people-related factors (Lee et al., 2023; Heimberger, Horvat and Schultmann, 2026). The present study builds on this literature but places particular emphasis on the transition from adoption to implementation and from implementation to sustainable value.

The chapter therefore explains the review philosophy, review design, literature identification and selection, analytical procedure, quality considerations, limitations and ethical position. The objective is not to claim statistical causality from the literature, but to produce a transparent and theoretically coherent synthesis of the evidence.

3.2 Research Philosophy

The study adopts an interpretive and pragmatic orientation appropriate to a meta-study of organisational AI research. A pragmatic element is necessary because the research question is applied: it seeks to explain what organisational conditions enable AI to create value. At the same time, the literature includes different epistemological positions, research designs and levels of analysis. Quantitative studies estimate relationships between AI capability and organisational outcomes, qualitative studies examine implementation processes and organisational experiences, while systematic reviews synthesise existing knowledge. Treating these forms of evidence as complementary allows the review to retain the richness of the literature rather than privileging a single methodological tradition.

The study therefore treats AI value realisation as a socio-technical and organisational phenomenon. Technology is considered necessary in many AI applications, but the literature indicates that value depends on how technological resources interact with people, processes, data, organisational structures and governance. Mikalef and Gupta (2021), for example, conceptualise AI capability as the mobilisation and orchestration of technological, human and organisational resources. More recent work similarly links AI capability to decision-making performance and organisational performance rather than treating AI adoption as an isolated technological variable (Emeagwali and Aljuhmani, 2024).

3.3 Research Approach

The research uses an integrative meta-study approach combining systematic literature identification, thematic synthesis and cross-study comparison. The term meta-study is used here in a broad methodological sense: the objective is to examine findings across a body of peer-reviewed research, compare the assumptions and explanations used by different studies, and develop an integrated interpretation of the conditions associated with AI value creation. This is distinct from a statistical meta-analysis, which would require sufficiently comparable quantitative effect sizes across studies.

The distinction is important because the recent AI literature is methodologically heterogeneous. It includes systematic reviews, surveys, structural equation modelling, field experiments, qualitative case studies and conceptual work. A statistical aggregation of all such evidence would risk obscuring important differences in constructs, contexts and outcomes. The meta-study instead asks what patterns recur across studies and under what conditions apparently conflicting findings can be reconciled.

This approach is consistent with recent review research. Lee et al. (2023) systematically categorise AI implementation research into organisational, information-systems, technological and people dimensions. Heimberger, Horvat and Schultmann (2026) review drivers of AI adoption in production, while a 2024 systematic review of organisational AI adoption and integration identifies technological, organisational and environmental factors as interconnected drivers and barriers. These reviews provide methodological precedent for synthesising heterogeneous evidence rather than treating individual studies as isolated findings.

3.4 Review Design

The review is designed as a structured integrative literature review with a meta-study orientation. The review focuses on peer-reviewed literature published primarily between 2021 and 2026, while retaining earlier foundational work where it is directly necessary to explain concepts that continue to structure the field. The time window is intended to capture the transition from conventional AI and machine learning research towards the organisational implications of generative and increasingly autonomous AI.

The review is organised around the following analytical dimensions:

AI adoption and AI capability: how organisations develop the resources and capabilities needed to use AI.

Strategic alignment and operating models: how AI initiatives are connected to business priorities, ownership and organisational structures.

Implementation and scaling: how organisations move from experimentation and proof of concept to production and wider deployment.

Data, systems and business-process integration: how data quality, interoperability and process design condition AI value.

People and human–AI collaboration: how employee skills, trust, participation and task characteristics influence outcomes.

Governance and responsible AI: how organisations establish accountability, risk controls, transparency and human oversight.

Business value and sustainability: how operational improvements develop into organisational, commercial and strategic outcomes.

3.5 Literature Identification Strategy

The literature search is conceptually structured around combinations of the following search themes: artificial intelligence and organisational adoption; AI capability and firm performance; AI implementation and scaling; AI business value; AI and business-process performance; human–AI collaboration; responsible AI governance; and organisational readiness. The principal scholarly sources considered are peer-reviewed journals and established academic publishers in information systems, management, operations, organisational behaviour and technology management.

Priority is given to studies that directly address organisational AI rather than purely technical model performance. Studies are particularly relevant where they examine organisational capabilities, implementation conditions, performance outcomes, governance, employee effects or mechanisms connecting AI use to value. Systematic and meta-analytic studies are given additional analytical weight because they themselves synthesise prior evidence.

The resulting corpus is therefore not intended to represent every publication on artificial intelligence. It is a theoretically focused corpus designed to answer the research question. This is consistent with the review logic proposed by Snyder (2019), under which literature reviews should be structured around the research purpose and transparent inclusion logic rather than treated as undifferentiated collections of sources.

3.6 Inclusion and Exclusion Criteria

Studies were considered relevant where they met most or all of the following criteria: (1) peer-reviewed publication; (2) organisational or workplace context; (3) explicit consideration of AI adoption, capability, implementation, governance, human–AI interaction or business value; (4) sufficient methodological or conceptual detail to support evaluation; and (5) publication within the principal 2021–2026 review window or clear foundational importance.

Purely technical studies focused on model architecture, benchmark performance or algorithmic optimisation without organisational implications were excluded. Industry reports, consultancy reports and non-peer-reviewed commentary were also excluded from the evidential core. They may be acknowledged as contextual material elsewhere in the dissertation, but they do not constitute evidence for the meta-study's conclusions. This distinction is important because the revised study is explicitly positioned as a literature-based academic investigation.

3.7 Analytical Procedure

The analysis proceeds in four stages. First, studies are mapped according to publication year, research design, organisational level and principal construct. Second, findings are coded into the analytical dimensions identified above. Third, recurring relationships are compared across studies, with attention to convergence, disagreement and boundary conditions. Fourth, the themes are synthesised into an integrated explanation of AI value realisation.

The coding is both deductive and inductive. Deductive categories are derived from the research question and the existing AI-capability literature. Inductive coding permits additional themes to emerge where the recent literature identifies issues not fully captured by the initial framework. Examples include the distinction between adoption and production-scale implementation, the importance of decision-making speed and quality, the unevenness of AI performance across tasks, and the organisational role of responsible-AI governance.

Particular attention is paid to mechanisms rather than simple associations. For example, the question is not merely whether AI is associated with performance, but whether the literature explains that relationship through decision-making, business-process performance, process innovation, organisational learning, employee augmentation or other complementary capabilities. Zebec and Indihar Štemberger (2024) provide particularly relevant evidence by showing how business-process-management capabilities can mediate the relationship between AI adoption and organisational performance.

3.8 Quality and Critical Appraisal

Quality appraisal focuses on methodological transparency, appropriateness of research design, clarity of constructs, contextual relevance and consistency between evidence and claims. Quantitative studies are considered in terms of sample, measurement and analytical approach; qualitative studies are assessed for contextual richness and transparency; systematic reviews are examined for the clarity of their search and synthesis procedures.

The review also distinguishes between evidence of association and evidence of causation. This is essential because organisational AI research includes both correlational studies and stronger experimental designs. For example, Dell’Acqua et al. (2026) use a field experiment to demonstrate that generative AI can improve performance on tasks within its capability frontier while harming performance on tasks outside that frontier. Such evidence is treated differently from cross-sectional survey evidence because it provides stronger grounds for causal interpretation within its experimental context.

Triangulation occurs conceptually across research designs. Where survey research indicates a positive relationship between AI capability and performance, the review examines whether qualitative studies and systematic reviews provide plausible organisational mechanisms explaining that relationship. Conversely, where research identifies productivity gains, studies reporting negative employment or task effects are considered so that the synthesis does not reduce AI value to a uniformly positive outcome.

3.9 Scope of the Meta-Study

The meta-study examines organisational AI value rather than technical AI performance. The unit of analysis is therefore the organisation, organisational function, work system or employee–AI relationship, depending on the study being synthesised. The review covers conventional AI, machine learning and generative AI where these technologies illuminate organisational capability and value realisation.

The study does not attempt to calculate a single universal return on AI investment. Nor does it assume that all AI applications generate the same type of value. Instead, value is treated as multidimensional, encompassing productivity, decision quality, process performance, innovation, customer outcomes, resilience and strategic transformation. This approach reflects the finding that AI effects vary by task, worker, technology and organisational context (Dell’Acqua et al., 2026; Ye et al., 2025).

3.10 Ethical Considerations

Because the study relies on published peer-reviewed literature rather than human participants or confidential organisational data, no primary participant recruitment or consent process is required. Ethical responsibility instead concerns accurate representation of published evidence, avoidance of selective citation and clear differentiation between authors' findings and the researcher's synthesis. Claims are therefore framed proportionately to the underlying evidence, and contradictory findings are retained where they are theoretically meaningful.

3.11 Methodological Limitations

The meta-study has several limitations. First, the AI literature is evolving rapidly, meaning that the evidence base can change as new organisational studies are published. Second, differences in constructs and outcome measures limit direct comparison between studies. Third, the literature contains publication and methodological biases, including an emphasis on organisations with sufficient digital maturity to adopt AI. Fourth, much of the recent literature remains cross-sectional, limiting conclusions about long-term value realisation.

A further limitation is that generative AI and agentic AI are relatively new. The strongest evidence on these technologies is therefore often task-specific or early-stage. Findings about productivity should not automatically be generalised to enterprise-level financial performance. Similarly, evidence about AI adoption should not be interpreted as proof that adoption itself causes sustainable competitive advantage.

These limitations reinforce the value of the meta-study. Rather than presenting a single deterministic model, the review identifies a pattern of complementary organisational capabilities and specifies where the evidence remains uncertain.

3.12 Chapter Summary

This chapter has reframed the research as a meta-study of recent peer-reviewed literature. The methodology combines structured literature identification, critical appraisal, thematic coding and cross-study synthesis. The approach is appropriate because AI value research is heterogeneous and because the central question concerns organisational mechanisms rather than a single measurable treatment effect. The next chapter presents the results of the literature synthesis.

4. Findings 

4.1 Introduction

This chapter presents the findings emerging from the synthesis of recent peer-reviewed literature (2021-2026). Seven themes dominate the evidence: AI adoption as capability development; strategic alignment and business ownership; the transition from pilot to production; data and process integration; employee enablement and human–AI collaboration; responsible AI governance; and the progression from efficiency to broader enterprise value.

4.2 AI Adoption Is Increasingly Conceptualised as Capability Development

The literature strongly challenges a simple technology-adoption view of AI. Mikalef and Gupta (2021) define AI capability through the mobilisation of technological, human and organisational resources and find positive relationships with organisational creativity and firm performance. Subsequent studies reinforce this capability perspective by showing that AI produces organisational outcomes through mediating mechanisms rather than through technology ownership alone.

The literature therefore distinguishes three analytically different stages. Adoption concerns whether AI is accepted or introduced. Implementation concerns whether it becomes embedded in organisational activities. Value realisation concerns whether that implementation generates measurable outcomes. Lee et al. (2023) show that implementation is influenced by organisational, information-system, technological and people dimensions, while Heimberger, Horvat and Schultmann (2026) identify multiple drivers and conditions affecting the transition into production.

The emerging consensus is that mature AI organisations are not simply those with more AI applications. They are those with repeatable capabilities for selecting valuable use cases, integrating data, managing implementation, developing employee skills, governing risk and measuring outcomes.

4.3 Strategic Alignment and Business Ownership

A second consistent theme is the importance of strategic alignment. AI value is greater when the technology is connected to a defined organisational problem or strategic objective. Lee et al. (2024), in their case-based analysis of AI orientation and capabilities, link strategic AI orientation with the development of capabilities that can contribute to business value. This supports the argument that AI capability is partly shaped by the organisation's broader strategic intent.

Recent quantitative research also connects AI capability to decision-making and organisational performance. Emeagwali and Aljuhmani (2024) find that AI capability influences decision-making speed and quality, which in turn contribute to organisational performance. The implication is that AI creates value when it improves organisational processes that matter to strategic execution.

The literature also points to the importance of business ownership. Technical specialists are essential for data, models, architecture and deployment, but business functions possess the contextual knowledge required to select meaningful use cases, redesign processes and assess commercial outcomes. The strongest interpretation is therefore not that AI should be owned by IT or by business units alone, but that AI requires cross-functional ownership.

4.4 The Pilot-to-Production Gap

The transition from experimentation to production is repeatedly identified as a major organisational challenge. Mikalef et al. (2021) demonstrate that AI adoption can affect performance through decision-making and business-process performance, while the implementation literature explains why such benefits are difficult to scale. Aydiner et al. (2022) emphasise AI-specific capabilities associated with planning, co-development, data management and lifecycle management.

Recent review research identifies productionisation and scaling as distinct stages rather than as automatic extensions of proof of concept (Heimberger, Horvat and Schultmann, 2026). This distinction is analytically important. A pilot can succeed under controlled conditions while remaining economically or organisationally unsuitable for enterprise deployment. Production requires reliable data pipelines, interoperability, security, governance, operating ownership, user adoption and sustainable cost structures.

The literature therefore supports interpreting the pilot-to-production gap as a capability gap. Technical feasibility is only one component of production readiness. Organisations need to design for scale from the beginning rather than treating scale as a later technical exercise.

4.5 Data, Systems and Business-Process Integration

Data quality and integration emerge as foundational conditions for AI value. Enholm et al. (2022) identify data availability, quality and infrastructure as important enablers of AI business value. More recent research strengthens this process-oriented interpretation. Zebec and Indihar Štemberger (2024) show that business-process-management capabilities can mediate the relationship between AI adoption and organisational performance, with process automation, organisational learning and process innovation acting as complementary mechanisms.

The synthesis therefore indicates that data should not be considered an isolated technical input. Data creates value when it is connected to systems and processes through which decisions and actions occur. A model may generate an accurate prediction, but the prediction has limited organisational value if employees cannot access it at the relevant point in the workflow or if no process exists for acting on it.

This finding helps explain why AI value may remain localised. Optimising a single task can produce productivity gains without changing the performance of the wider process. Enterprise value requires integration across activities, functions and decision points.

4.6 Organisational Readiness and Human–AI Collaboration

The literature provides strong evidence that AI value is mediated by human capabilities. Mikalef and Gupta (2021) place human skills alongside technological and organisational resources in the construction of AI capability. More recent work shows that the fit between organisational AI adoption and employee AI skills matters for organisational performance, with employee–AI collaboration providing an important mediating mechanism (Zhao et al., 2025).

Generative AI research further demonstrates that the effects of AI are not uniform. Brynjolfsson, Li and Raymond (2025) find substantial productivity improvements in customer support, particularly among less experienced workers. Dell’Acqua et al. (2026), however, show that AI performance is uneven across tasks: users performed better and faster on tasks within the technology's capability frontier but performed worse on a task outside that frontier.

The combined evidence suggests that AI literacy should involve more than tool operation. Employees need to understand when AI is reliable, when outputs require verification and when human judgement should override the system. AI adoption is therefore partly a work-design problem involving task allocation, decision rights, training, trust and workflow redesign.

The literature also warns against an exclusively automation-centred interpretation. Bankins et al. (2023) show that human–AI collaboration, attitudes towards AI, algorithmic management and labour-market implications are interconnected. AI may augment human capability in some contexts while substituting for work or intensifying control in others. Sustainable enterprise value therefore requires deliberate design of the human–AI relationship.

4.7 Responsible AI Governance

Governance is increasingly treated in the literature as an organisational capability rather than a narrow compliance function. Responsible AI governance concerns accountability, risk assessment, data use, transparency, monitoring, security and human oversight. The recent review by Papagiannidis et al. identifies the need to distinguish ethical principles from the organisational governance mechanisms through which those principles are implemented (Papagiannidis et al., 2025).

Sayan and Buechl (2025) similarly demonstrate that internal AI governance requires strategic direction, operational execution and coordination across organisational functions. Research on algorithm review boards also highlights the importance of leadership support and integration with existing organisational processes (Young et al., 2024).

The synthesis therefore suggests that governance should be embedded across the AI lifecycle. Use-case selection, data preparation, development, deployment, monitoring and retirement each involve different risks. Proportionate governance is particularly important: high-risk applications require stronger controls, while low-risk experimentation should not be unnecessarily constrained.

Governance can consequently support value creation in two ways. First, it reduces exposure to legal, operational, security and reputational risks. Second, clear governance can increase organisational confidence in AI, making responsible scaling easier.

4.8 From Efficiency to Enterprise Value

The literature consistently distinguishes immediate operational effects from broader organisational value. Enholm et al. (2022) distinguish first-order effects, such as faster task completion and reduced errors, from second-order effects, such as profitability, innovation and competitive advantage. This distinction provides a useful explanation for why productivity benefits can emerge before strategic transformation.

Recent empirical evidence supports the existence of productivity effects but also shows that they are conditional. Brynjolfsson, Li and Raymond (2025) demonstrate substantial productivity benefits in customer support. Dell’Acqua et al. (2026) show that benefits depend on task characteristics. A 2025 meta-analysis of enterprise AI research likewise reports a stronger relationship between AI adoption and innovation than between AI adoption and organisational performance, while identifying substantial moderators including technology type, industry and human–AI interaction (Ye et al., 2025).

The synthesis therefore rejects the assumption that efficiency gains automatically become revenue or sustainable competitive advantage. Value requires organisations to redirect the capacity created by AI towards higher-value activities such as customer experience, innovation, product development, strategic analysis and improved decision-making.

4.9 Cross-Study Synthesis

Across the literature, the strongest recurring pattern is complementarity. Strategy without operating mechanisms produces aspiration rather than execution. AI capability without employee skills limits adoption. Data without integration produces technically useful but operationally disconnected outputs. Governance without business integration can become a compliance exercise, while rapid adoption without governance can undermine trust and sustainability.

The literature can therefore be represented as a capability chain: strategic intent shapes use-case selection; operating models allocate ownership and resources; data and systems enable implementation; people and process redesign embed AI in work; governance creates accountable and trusted use; and measurement determines whether the resulting changes constitute business value.

This is not a linear causal sequence. The capabilities reinforce one another. Weakness at any point can constrain the value generated elsewhere. The evidence therefore supports a system-level interpretation of AI value realisation rather than a single-variable adoption model.

4.10 Chapter Summary

The meta-study finds that the organisational value of AI depends less on adoption itself than on the complementary capabilities surrounding adoption. The most consistent themes are strategic alignment, cross-functional ownership, production-scale implementation, data and process integration, employee capability, human–AI collaboration, responsible governance and deliberate conversion of productivity into broader value. These findings provide the basis for the discussion in Chapter 5.

5. Discussion

5.1 Introduction

The meta-study demonstrates that the central problem in AI-enabled transformation is not technological availability but organisational realisation. Across systematic reviews, quantitative studies, qualitative research and field experiments, the literature repeatedly indicates that AI becomes economically meaningful when organisations develop the capabilities required to integrate it into work, decisions and business processes.

The discussion develops this argument through five propositions: AI adoption is a capability-development process; strategic alignment and operating design mediate value; scaling depends on integration; human–AI collaboration is a source of both value and risk; and sustainable value requires movement from first-order efficiency to second-order organisational and strategic outcomes.

5.2 AI Adoption as an Organisational Capability

The first proposition is that AI adoption should be understood as capability development. This extends the resource-based view used by Mikalef and Gupta (2021). AI resources are not valuable in isolation. Their value depends on the organisation's ability to combine technical resources with data, human expertise, managerial support, organisational routines and processes.

This capability interpretation is strengthened by the more recent literature. Emeagwali and Aljuhmani (2024) demonstrate that AI capability affects decision-making speed and quality, while Lee et al. (2024) connect AI orientation with the development of AI capabilities and process-oriented dynamic capabilities. The implication is that organisational maturity should be assessed through repeatable capabilities rather than through the number of AI tools deployed.

This also explains why AI adoption can coexist with weak business performance. An organisation may possess substantial technical capability but lack the complementary structures required to turn that capability into improved processes, customer outcomes or strategic differentiation.

5.3 Strategic Alignment and the AI Operating Model

The second proposition is that strategic alignment and operating design form the bridge between technological possibility and organisational execution. AI initiatives are more likely to create value when they begin with a business problem, have an accountable owner and have measurable outcomes.

The literature suggests that an AI operating model should allocate decision rights across business, technology, data and governance functions. Centralised technical expertise can provide consistency and economies of scale, while business ownership ensures that use cases remain connected to customer, operational and commercial priorities. A hybrid model is therefore often more plausible than either complete centralisation or complete decentralisation.

The meta-study also reveals that alignment is dynamic. As AI capabilities evolve, strategic priorities may change and previously infeasible use cases may become viable. The operating model must therefore support portfolio learning rather than merely project execution.

5.4 The Pilot-to-Production Problem as an Organisational Capability Gap

The pilot-to-production problem is best interpreted as a mismatch between experimental capability and operational capability. Experiments reward speed, technical learning and narrow problem definition. Production requires reliability, integration, security, governance, ownership, support and cost discipline.

The scaling literature supports a staged model in which organisations move from proof of concept to productionisation and then to broader platform or portfolio deployment (Heimberger, Horvat and Schultmann, 2026). This implies that scaling criteria should be defined before pilots begin. Data readiness, integration requirements, governance, employee adoption and total cost should form part of the initial business case.

This interpretation also explains why organisations can experience a high volume of successful pilots without corresponding enterprise transformation. The limiting factor is not necessarily innovation capacity; it may be the organisation's ability to institutionalise and repeat what works.

5.5 Data and Process Integration as Foundations of Value

The literature provides strong support for viewing data and integration as strategic infrastructure. AI systems are dependent on the quality, accessibility and context of data, but enterprise value depends additionally on whether outputs can influence operational processes.

Zebec and Indihar Štemberger (2024) are particularly important in this regard because their findings position business-process-management capabilities as mechanisms through which AI adoption translates into organisational performance. This shifts attention from model performance to workflow performance. The relevant question becomes not 'How accurate is the model?' but 'Does the AI-enabled process perform better, and does that improvement matter to the organisation?'

This distinction has practical consequences. Organisations may need to simplify or redesign processes before automating them. Automating a fragmented process can reproduce inefficiency at greater speed. Integration and process redesign should therefore precede large-scale automation where the underlying workflow is structurally weak.

5.6 Employee Enablement and Human–AI Collaboration

The evidence supports a human-centred interpretation of AI capability. Employee skills, trust and participation determine whether AI is embedded into everyday work. The relationship is reciprocal: employees influence how AI is used, while AI changes tasks, roles and decision-making.

The productivity literature illustrates both the opportunity and the boundary condition. Brynjolfsson, Li and Raymond (2025) show that AI assistance can raise productivity and disproportionately benefit less experienced workers, suggesting that AI can diffuse expertise. Dell’Acqua et al. (2026), however, demonstrate that the same technology can reduce performance on tasks outside its capability frontier. Effective AI implementation therefore requires task-level understanding and appropriate human oversight.

The implication is that AI literacy must include critical judgement. Employees need to know when to accept, verify, challenge or reject AI output. Organisational learning mechanisms are equally important because successful practices must be shared across functions if value is to scale.

5.7 Governance as an Enabler of Sustainable AI

Responsible governance should be understood as part of the value architecture rather than an external constraint. The literature indicates that governance becomes more complex as AI moves from advisory applications to integrated and autonomous systems.

Sayan and Buechl (2025) show that effective governance is multi-level and cross-functional. The responsible-AI governance literature similarly emphasises the organisational translation of principles into operational controls. Governance therefore needs to be connected to actual workflows, decision rights and accountability.

A risk-proportionate model is preferable to either unrestricted experimentation or universal control. High-impact applications should require stronger assessment, monitoring and human oversight. Lower-risk applications can operate within simpler controls. This approach allows governance to support experimentation while protecting the organisation and its stakeholders.

5.8 Why Efficiency Often Precedes Strategic Value

The distinction between first-order and second-order effects provides the strongest explanation for the common sequencing of AI value. First-order improvements are close to the task affected by the technology and are therefore easier to observe. Second-order outcomes require organisational changes beyond the initial AI application.

An AI tool may reduce processing time, for example, but sustainable enterprise value arises only if the organisation uses the released capacity to improve service, increase sales, develop products or strengthen strategic decision-making. The transformation therefore has an organisational allocation problem: who decides where the productivity dividend goes?

This explains why a productivity-focused AI strategy can become self-limiting. Efficiency is valuable, but if the organisation simply absorbs the time saving without redesigning work, the technology may not generate significant strategic differentiation.

5.9 From Productivity to Strategic Transformation

The meta-study suggests a three-stage value progression. The first stage is task-level augmentation or automation. The second is process-level redesign and integration. The third is strategic transformation, in which AI enables new products, services, customer experiences, decision systems or business models.

The stages should not be treated as universally sequential, because some organisations may develop strategic applications directly. Nevertheless, the distinction is analytically useful. It prevents early productivity gains from being confused with complete transformation and directs attention towards the organisational changes needed for higher-order value.

The evidence also suggests that innovation effects may be more robust than immediate financial effects. Ye et al. (2025) find a moderately strong positive relationship between AI adoption and enterprise and employee innovation, while the relationship with performance is weaker. This indicates that the value of AI may initially appear as increased organisational possibility rather than immediately observable financial returns.

5.10 An Integrated Model of AI Value Realisation

The literature can be synthesised into an integrated model comprising six mutually reinforcing capabilities:

Strategic alignment — defining why AI matters and selecting use cases connected to strategic priorities.

Operating-model capability — allocating ownership, decision rights, resources and coordination mechanisms.

Data and integration capability — ensuring reliable data, interoperable systems and workflow connectivity.

Human and organisational capability — developing AI literacy, trust, participation and organisational learning.

Governance capability — embedding accountability, risk management, monitoring and human oversight.

Value-management capability — measuring operational, customer, financial, innovation and strategic outcomes.

These capabilities interact rather than operate independently. Strategic alignment determines which capabilities should be built; the operating model determines how they are coordinated; data and integration make AI operationally possible; people make it usable; governance makes it trustworthy; and value management determines whether the overall system is producing the intended outcomes.

5.11 Implications for Theory

The principal theoretical implication is that AI business value should be conceptualised as a system-level organisational outcome. The resource-based perspective remains useful, but it should be extended towards a dynamic capability interpretation in which organisations continually sense opportunities, integrate AI resources, redesign processes and reconfigure work.

Second, adoption, implementation and value realisation should remain analytically distinct. Conflating these stages obscures the mechanisms through which technology affects performance.

Third, the evidence strengthens a socio-technical view of AI. Technology, people, data, processes and governance are mutually dependent. Finally, the literature suggests that organisational maturity is a boundary condition: the same AI technology can produce different outcomes depending on the capabilities surrounding it.

5.12 Implications for Management Practice

Managers should evaluate AI portfolios through business outcomes rather than technology counts. Every major initiative should have a business problem, accountable owner, scaling hypothesis, integration assessment, governance classification and value-measurement plan.

Organisations should also invest in reusable infrastructure and organisational learning. Data standards, integration patterns, governance processes, training and reusable AI components can reduce the marginal cost of subsequent AI initiatives. Finally, productivity gains should be explicitly linked to reinvestment decisions so that the AI dividend contributes to innovation and growth rather than becoming only a short-term cost reduction.

5.13 Answer to the Research Question

The research question asks how organisations can develop the capabilities required to convert AI adoption into sustainable enterprise value. The meta-study indicates that they do so by building an integrated organisational capability system rather than by increasing adoption alone. Strategic alignment identifies valuable opportunities; an operating model converts strategy into accountable execution; data and process integration embed AI into workflows; employee capability enables effective human–AI collaboration; responsible governance creates trust and control; and value management converts operational improvements into measurable organisational and strategic outcomes.

The central answer is therefore capability-based: sustainable value arises when organisations can repeatedly select, integrate, govern, scale and improve AI applications in ways that change how work is performed and how value is created.

6. Conclusions and Recommendations

6.1 Introduction

This chapter consolidates the conclusions of the meta-study and translates them into recommendations for organisations. Because the study is literature-based, the conclusions refer to the weight and convergence of peer-reviewed evidence rather than to findings from a single organisation or survey.

6.2 Main Conclusions

First, AI adoption is not equivalent to AI value. The literature consistently distinguishes between technological uptake and organisational realisation. Adoption creates potential; capability determines whether that potential is converted into outcomes.

Second, strategic alignment and operating design are central. AI initiatives need business ownership, clear decision rights and mechanisms for moving from use-case selection to production and measurement.

Third, data and integration are foundational. AI value depends on reliable information and the ability to connect AI outputs with the systems and processes through which organisations act.

Fourth, employees remain central. Human–AI collaboration can improve productivity and decision quality, but effects vary across tasks and workers. Organisations therefore require AI literacy, task-level judgement and mechanisms for human oversight.

Fifth, governance is a condition for sustainable scaling. Responsible AI governance must operate across the lifecycle and across organisational functions.

Sixth, efficiency is often an early value outcome rather than the endpoint. Broader strategic value requires process redesign and deliberate reinvestment of AI-generated capacity.

Finally, the evidence suggests complementarity. Sustainable value depends on the joint development of strategy, operating models, data, processes, people, governance and measurement.

6.3 Recommendations for Organisations

Recommendation 1: Link AI Investment to Strategic Problems

Every significant AI initiative should begin with a defined business problem, strategic rationale, expected outcome and accountable owner. Technology availability should not be treated as sufficient justification for investment.

Recommendation 2: Design for Production from the Start

Pilot proposals should include a credible path to production, covering data, integration, security, governance, operating ownership, user adoption and cost. Scaling criteria should be established before technical experimentation begins.

Recommendation 3: Treat Data and Integration as Strategic Infrastructure

Organisations should prioritise data quality, interoperability, access controls, metadata and integration architecture. Foundational remediation should be considered part of AI transformation rather than a separate technical programme.

Recommendation 4: Build Organisation-Wide AI Literacy

AI training should extend beyond specialists. Managers and employees should understand capabilities, limitations, verification requirements and appropriate escalation. Communities of practice and AI champions can support organisational learning.

Recommendation 5: Embed Responsible Governance

Governance should begin at use-case selection and continue through development, deployment, monitoring and retirement. Controls should be proportionate to risk and should include accountability, privacy, security, human oversight and performance monitoring.

Recommendation 6: Measure Value Beyond Productivity

Value measurement should include operational efficiency but also decision quality, customer outcomes, innovation, revenue, resilience, risk and strategic contribution. This enables management to determine whether early productivity gains are becoming broader enterprise value.

6.4 Limitations of the Meta-Study

The principal limitation is the heterogeneity of the literature. Studies differ in definitions of AI, measures of performance, industries, countries, organisational levels and research methods. Consequently, the synthesis should not be interpreted as a universal causal model.

A second limitation is recency bias. Because generative AI and agentic AI are developing rapidly, the evidence base is changing. Third, much organisational research remains cross-sectional, making long-term value realisation difficult to establish. Fourth, positive AI outcomes may be more visible in organisations with stronger digital maturity, potentially creating selection effects in the literature.

Finally, the review does not provide a statistical estimate of the average effect of AI adoption on enterprise value. This is intentional: the underlying studies are too heterogeneous for a single effect-size estimate to capture the organisational mechanisms that are central to the research question.

6.5 Directions for Future Research

Future research should prioritise longitudinal studies following organisations from adoption through production and subsequent performance. Such work could establish whether productivity gains persist and whether they develop into innovation, revenue or competitive advantage.

Comparative research should examine alternative AI operating models, including centralised, federated and hybrid structures. Research should also investigate how governance maturity affects both innovation speed and organisational trust.

Further work is needed on human–AI collaboration, especially the task-level conditions under which augmentation outperforms automation. Finally, researchers should examine how organisations convert productivity gains into customer-facing and strategic value, rather than treating efficiency as the final outcome.

6.6 Final Conclusion

The literature indicates that the AI challenge is increasingly an execution challenge. Organisations can access and experiment with AI more easily than they can build the complementary capabilities required for sustainable value. The decisive advantage is therefore not simply technological possession, but organisational capacity to integrate, govern, learn and scale.

References

Aydiner, A.S., Tatoglu, E., Bayraktar, E., Zaim, S. and Delen, D. (2019) ‘Business analytics and firm performance: The mediating role of business process performance’, Journal of Business Research, 96, pp. 228–237.

Bankins, S., Formosa, P., Grigoropoulos, C. and K. (2023) ‘A multilevel review of artificial intelligence in organizations: Implications for organizational behavior research and practice’, Journal of Organizational Behavior.

Brynjolfsson, E., Li, D. and Raymond, L.R. (2025) ‘Generative AI at work’, The Quarterly Journal of Economics, 140(2), pp. 889–942.

Dell’Acqua, F., McFowland III, E., Mollick, E., Lifshitz-Assaf, H., Kellogg, K.C., Rajendran, S., Krayer, L., Candelon, F. and Lakhani, K.R. (2026) ‘Navigating the jagged technological frontier: Field experimental evidence of the effects of artificial intelligence on knowledge worker productivity and quality’, Organization Science, 37(2).

Emeagwali, O.L. and Aljuhmani, H.Y. (2024) ‘Artificial intelligence capability and organizational performance: Unraveling the mediating mechanisms of decision-making processes’, Management Decision, 63(10), pp. 3501–3532.

Enholm, I.M., Papagiannidis, E., Mikalef, P. and Krogstie, J. (2022) ‘Artificial intelligence and business value: A literature review’, Information Systems Frontiers, 24(5), pp. 1709–1734. doi: 10.1007/s10796-021-10186-w.

Heimberger, H., Horvat, D. and Schultmann, F. (2026) ‘Exploring the factors driving AI adoption in production: A systematic literature review and future research agenda’, Information Technology and Management, 27, pp. 53–69.

Lee, M.C.M., Scheepers, H., Lui, A.K.H. and Ngai, E.W.T. (2023) ‘The implementation of artificial intelligence in organizations: A systematic literature review’, Information & Management, 60(5), 103816. doi: 10.1016/j.im.2023.103816.

Mikalef, P. and Gupta, M. (2021) ‘Artificial intelligence capability: Conceptualization, measurement calibration, and empirical study on its impact on organizational creativity and firm performance’, Information & Management, 58(3), 103434.

Mikalef, P., Islam, N., Parida, V., Singh, H. and Wastell, D. (2021) ‘Artificial intelligence (AI) competencies and firm performance: The mediating role of AI-based process innovation and the moderating role of digital maturity’, Information Systems Frontiers.

OMMAX, Ibexa and Make (2026) AI Trends Report 2026.

Papagiannidis Emmanouil,, Patrick Mikalef & Conboy Kieran, 2025, ‘Responsible artificial intelligence governance: A review and research framework’, Journal of Strategic Information Systems, 34(2), 101885. doi: 10.1016/j.jsis.2024.101885.

Sayan, M.M. and Buechl, J. (2025) ‘Embedding AI governance in organizations: A multi-level approach to responsible AI use’, Procedia Computer Science, 270, pp. 4144–4154.

Snyder, H. (2019) ‘Literature review as a research methodology: An overview and guidelines’, Journal of Business Research, 104, pp. 333–339.

Weber, M., Engert, M., Schaffer, N., Weking, J. and Krcmar, H. (2022) ‘Organizational capabilities for AI implementation—Coping with inscrutability and data dependency in AI’, Information Systems Frontiers.

Ye, Z., Chen, X., Mai, Y. and Duan, C. (2025) ‘The effectiveness and boundary conditions of AI adoption in enterprise: A meta-analysis study’, Academy of Management Proceedings, 2025(1). doi: 10.5465/AMPROC.2025.19698.

Young, M., Katell, M. and Krafft, P. (2024) ‘Investigating algorithm review boards for organizational responsible artificial intelligence governance’, AI and Ethics.

Zebec, A. and Indihar Štemberger, M. (2024) ‘Creating AI business value through BPM capabilities’, Business Process Management Journal, 30(8), pp. 1–26. doi: 10.1108/BPMJ-07-2023-0566.

Zhao, R., Niu, L., Chen, Z. and Fan, Y. (2025) ‘How organizational AI adoption and employee AI skills fit to affect organizational performance’, Academy of Management Annual Meeting Proceedings, 2025(1). doi: 10.5465/AMPROC.2025.12059.

Contact

Reach out via email for inquiries.

Email

Subscribe to newsletter

info@grcadvisory.ch

© 2025. All rights reserved.