The AI-Enabled CFO: Redesigning Finance for Sustainable Competitive Advantage

Abstract

Artificial intelligence is increasingly transforming the finance function from a primarily information-processing activity into a technology-enabled decision-support capability. Yet the organizational conditions under which AI generates sustainable value remain insufficiently understood. This paper develops a CFO-oriented framework for AI transformation by integrating the practitioner perspective of Parnova Consulting's CFO AI Navigator 2026 with recent peer-reviewed research on artificial intelligence, accounting, management accounting, governance, and knowledge-work productivity. Drawing on evidence from experimental and field studies, literature reviews, and research frameworks, the paper argues that the economic value of AI in finance cannot be inferred from technological capability or task-level productivity improvements alone. Instead, value depends on the interaction among AI capabilities, process maturity, data quality, human expertise, governance, and organizational decision-making.

The paper proposes that the appropriate unit of analysis is the AI-enabled finance process, rather than the individual AI tool. It develops a five-stage CFO transformation model comprising diagnosis, prioritization, experimentation, process redesign, and institutionalization. The analysis further identifies six propositions concerning CFO ownership, process-based use-case selection, data and process maturity, human-AI complementarity, governance maturity, and the shift from information production toward decision enablement. The paper argues that AI should be managed as a portfolio of investments and evaluated using both operational and decision-quality outcomes rather than generic productivity or return-on-investment claims.

The central conclusion is that successful finance AI transformation is not primarily a technology-adoption problem. It is an organizational redesign challenge in which technology, processes, people, data, controls, and decision rights must be deliberately integrated. For CFOs and boards, the strategic objective should therefore not be maximum automation, but the creation of a more productive, analytically capable, governable, and strategically relevant finance function.

Keywords: artificial intelligence; finance function; CFO; management accounting; generative AI; human-AI collaboration; AI governance; digital transformation; finance transformation; organizational capability

1. Introduction

Artificial intelligence (AI) has rapidly evolved from an emerging digital technology into an organizational capability with potentially profound implications for how firms generate information, make decisions, coordinate activities, and exercise control. This transformation is particularly consequential for corporate finance. Finance functions combine large volumes of structured and unstructured data with highly repetitive processes, forecasting and planning activities, regulatory and control requirements, and decisions that depend on professional judgment. Consequently, activities such as accounts payable and receivable, financial close, management reporting, forecasting, liquidity management, anomaly detection, compliance, and financial analysis constitute important domains for AI-enabled process redesign. The strategic significance of AI in finance therefore extends beyond automating individual tasks: it concerns how the finance function itself is organized, how information is produced and interpreted, and how finance professionals contribute to managerial decision-making.

Recent empirical research provides compelling evidence of AI's productivity potential, but also cautions against viewing these gains as automatic or universal. Noy and Zhang (2023), in a randomized experiment involving professional writing tasks, find that generative AI substantially reduces task completion time while improving average output quality. Their results demonstrate that generative AI can produce significant productivity gains in knowledge-intensive work, although the experiment also highlights the importance of task characteristics and user interaction. Brynjolfsson, Li, and Raymond (2025) provide complementary field evidence from 5,172 customer-support workers: access to a generative-AI assistant increased productivity, measured by issues resolved per hour, by approximately 15% on average. Crucially, however, the effects were heterogeneous. Less-experienced and lower-skilled workers experienced larger improvements in both speed and quality, whereas the most experienced workers experienced smaller gains in speed and modest declines in quality. These findings challenge a simple automation narrative and instead point toward AI as a technology whose value depends on the interaction between task characteristics, human expertise, and organizational context.

This heterogeneity is further reinforced by Dell'Acqua et al. (2026), whose field experiment demonstrates that AI performance is shaped by a "jagged technological frontier": AI can improve knowledge-worker performance substantially on tasks within its capability frontier, while potentially reducing performance on tasks that fall outside it. For finance functions, this distinction is critical. AI may be highly effective in extracting information from invoices, classifying transactions, summarizing financial documents, identifying anomalies, or generating initial analyses, while being less reliable when tasks require contextual interpretation, unusual accounting judgments, or complex organizational knowledge. AI adoption should therefore not be premised on the assumption that higher levels of automation necessarily produce higher levels of organizational performance. Rather, the relevant managerial question is where and under what conditions AI can complement human expertise.

The emerging accounting and finance literature provides further support for this more nuanced perspective. Abbas (2026), in a systematic review of 91 studies, shows that AI, machine learning, deep learning, explainable AI, generative AI, and large language models are transforming not only accounting technologies and information processes but also organizational structures, professional boundaries, and the role and skill requirements of management accountants. The review identifies four interconnected dimensions of transformation: digitalization of management accounting, adoption of AI technologies, changes in strategy and control and the transformation of accounting and finance functions, and the emergence of new roles and competencies for controllers and management accountants. At the same time, Abbas emphasizes unresolved concerns surrounding data privacy, confidentiality, security, ethics, trust, accuracy, explainability, and the continuing role of human judgment. This suggests that the consequences of AI adoption should be assessed at the level of the finance function and its organizational relationships rather than at the level of isolated technological applications.

The literature on generative AI reaches a similar conclusion. Dong, Stratopoulos, and Wang (2024), in their scoping review of research on ChatGPT and large language models in accounting and finance, identify three broad research streams. The first concerns applications of LLMs across accounting and finance domains; the second examines LLMs as tools for classification, summarization, text generation, and other research activities; and the third addresses the implications of LLM adoption for accounting and finance professionals, organizations, and sectors. Importantly, their review also distinguishes between conceptual research, proposed applications, case studies, and emerging evidence of value realization. This distinction highlights a persistent gap between technological possibility and demonstrated organizational value.

Stratopoulos and Wang (2025) extend this research agenda by emphasizing the implications of AI for accounting research and practice. Their framework reinforces the need to examine AI not simply as a new analytical instrument but as a technology capable of altering the production, interpretation, and use of accounting information. Taken together with Abbas (2026), this literature suggests that the adoption of AI can reshape the boundary between accounting, information systems, analytics, and managerial decision-making. The future finance professional is consequently unlikely to be defined solely by traditional accounting expertise; rather, the profession increasingly requires combinations of accounting knowledge, data literacy, technological competence, critical evaluation, and strategic business partnering.

This transformation also creates a governance challenge. AI-enabled finance processes operate on sensitive financial, employee, customer, and strategic information and may influence decisions with material financial and regulatory consequences. Eisikovits, Johnson, and Markelevich (2024) highlight the risks and opportunities associated with incorporating AI into accounting and auditing, including concerns relating to reliability, confidentiality, privacy, accountability, professional judgment, and the potential for inappropriate reliance on AI-generated outputs. These concerns become particularly important as organizations move from AI systems that merely assist employees toward systems capable of recommending, coordinating, or executing actions. The governance challenge is therefore not simply whether an AI model is technically accurate, but whether its use is appropriately controlled within the organization's broader system of accountability.

Recent research on AI governance reinforces this organizational perspective. Almeida and Santos Júnior (2025) show that effective AI governance involves multiple organizational layers and depends on capabilities such as training, organizational processes, and institutional support rather than on technical controls alone. Hadley, Blatecky, and Comfort (2025) similarly demonstrate the importance of organizational structures for responsible AI governance and emphasize the role of leadership support and integration with existing organizational processes. These findings are particularly relevant to finance because governance, internal control, risk management, and accountability are already central elements of the CFO's mandate. AI governance should therefore be understood not as an external compliance layer added to finance transformation, but as part of the finance function's evolving control architecture.

Against this academic background, the practitioner-oriented CFO AI Navigator 2026 developed by Parnova Consulting provides a complementary implementation perspective. The guide argues that AI should be treated as a strategic finance and leadership issue rather than as a conventional IT project. It recommends that CFOs assess organizational and process readiness, identify and prioritize a limited number of high-value use cases, establish baseline measures before implementation, define explicit success criteria, integrate change management and skills development, and introduce AI-specific governance—including considerations of data privacy, model validation, and explainability—from the outset (Parnova Consulting, 2026). Its emphasis on CFO ownership is particularly significant because AI can alter not only operational efficiency but also forecasting, management reporting, internal controls, decision rights, organizational capabilities, and resource allocation.

The Parnova framework also highlights a practical tension that is increasingly visible in the academic literature: the difference between AI experimentation and AI value realization. Organizations can readily generate a large portfolio of potential AI applications, but technological availability does not establish an economic business case. The productivity evidence from Noy and Zhang (2023), Brynjolfsson et al. (2025), and Dell'Acqua et al. (2026) indicates that AI effects vary according to tasks and users. Similarly, the accounting literature remains relatively limited in its empirical evidence concerning the organizational consequences of newer AI technologies (Abbas, 2026; Dong et al., 2024). For CFOs, this implies that AI investments should be evaluated through measurable changes in process cost, cycle time, error rates, forecast accuracy, decision quality, control effectiveness, and employee capability rather than through the number of pilots launched or the sophistication of the technology deployed.

The central implication is that the appropriate unit of analysis is not the AI tool, but the AI-enabled finance process. A tool may be technically impressive yet economically insignificant if the underlying process is fragmented, poorly governed, data-deficient, or organizationally unsuitable. Conversely, relatively mature AI capabilities can generate substantial value when they are embedded in well-designed processes with reliable data, clearly defined decision rights, appropriate human oversight, and measurable outcomes. AI value therefore emerges from the interaction of technology with organizational design rather than from technology in isolation.

This perspective also reframes the role of the CFO. CFO leadership does not imply that finance executives must become AI engineers. Rather, it means that they must determine where AI can create economic and strategic value, which decisions can appropriately be augmented or automated, what risks are acceptable, what controls are required, and how organizational capabilities should evolve. In this sense, the CFO becomes an orchestrator of human, technological, informational, and organizational resources. The transformation of finance is consequently less a question of replacing finance professionals than of reallocating their time and expertise from repetitive information production toward interpretation, challenge, scenario analysis, business partnering, and strategic decision support.

Building on these insights, this paper asks:

How should CFOs design and govern AI adoption so that technological capabilities translate into sustainable improvements in finance-function performance and decision quality?

The paper argues that effective AI adoption in finance requires a shift from technology-centred implementation to process-centred transformation. Sustainable value creation depends on the alignment of five elements: technology, data, processes, people, and governance. AI should therefore be introduced selectively, measured against credible baselines, deployed within a human-AI operating model, and governed as an organizational capability. This argument extends the practical propositions of the CFO AI Navigator 2026 by grounding them in the emerging empirical and conceptual literature on AI productivity, accounting transformation, professional roles, and responsible AI governance.

The remainder of the paper develops this argument by examining AI's implications for finance processes, the economics of use-case selection, human-AI complementarity, organizational and professional transformation, and AI governance. It then proposes a CFO-centred framework for moving from experimentation to scalable AI-enabled finance transformation and identifies directions for future empirical research.

2. AI in Finance: From Automation to Organizational Transformation

The economic significance of artificial intelligence (AI) in finance cannot be understood adequately through the conventional lens of automation. Traditional automation primarily substitutes technology for human effort in well-defined, repetitive, and rules-based activities. AI expands this logic by enabling systems to perform tasks that previously required elements of perception, classification, prediction, language processing, pattern recognition, and, increasingly, interactive reasoning. Generative AI and large language models (LLMs) extend these capabilities further by producing and transforming text, interpreting unstructured information, summarizing documents, generating analyses, and interacting with users in natural language. The resulting transformation is therefore not simply one of automating existing work, but of changing which activities are performed by humans and machines, how information moves through the organization, and where professional judgment is applied.

This distinction is particularly important for corporate finance. A conventional rules-based system may automate invoice entry or reconciliation once the relevant rules have been specified. Machine-learning systems can extend this capability by identifying anomalous transactions, classifying documents, predicting payment behaviour, or detecting patterns that are difficult to capture through deterministic rules. Generative AI can operate at a different level by interpreting financial and non-financial documents, producing preliminary management commentary, answering questions about financial information, supporting analysis, and translating complex information into different forms for different users. Emerging agentic architectures potentially extend the logic further by coordinating sequences of activities across systems, subject to specified constraints and human oversight. The technological progression can therefore be conceptualized as a movement from task automation, through decision augmentation, toward process orchestration.

The distinction matters because the economic value of AI is unlikely to arise uniformly across these three levels. At the automation level, value can often be captured through reductions in processing time, manual effort, error rates, or transaction costs. At the augmentation level, value depends more heavily on whether AI improves the quality, speed, or consistency of professional judgments. At the orchestration level, the principal opportunity is potentially more fundamental: the redesign of an entire finance process around AI-enabled information flows and decision rights. The relevant management question consequently shifts from which tasks can AI perform? to how should finance processes be redesigned when AI can perform some tasks differently, faster, or at greater scale?

Recent empirical evidence supports this more differentiated view of AI's economic effects. Noy and Zhang (2023) demonstrate that generative AI can substantially increase productivity in knowledge-intensive tasks, but their experimental design also shows that the magnitude and direction of the effect depend on the nature of the task and the interaction between users and the technology. Brynjolfsson, Li, and Raymond (2025) reach a similar conclusion in a large field study of customer-support workers: generative AI increased average productivity by approximately 15%, but the gains were highly heterogeneous across workers. In particular, less-experienced workers benefited disproportionately, suggesting that AI can partially diffuse expertise and reduce performance differences within a workforce. These findings are important for finance because they imply that AI should not be evaluated simply in terms of aggregate labour substitution. Its value may instead arise from changing the distribution of expertise, accelerating the work of less-experienced employees, and enabling experienced professionals to concentrate on higher-value activities.

Dell'Acqua et al. (2025) provide an especially relevant qualification through their concept of a jagged technological frontier. Their field experiment demonstrates that generative AI can improve knowledge-worker performance on tasks within its capability frontier while potentially reducing performance on tasks that fall outside that frontier. This finding challenges the assumption that more AI necessarily produces better performance. For finance, the implication is substantial: an AI system may be highly effective at extracting information from invoices, summarizing financial reports, identifying anomalies, or producing a first-pass analysis, while remaining unreliable when confronted with unusual transactions, ambiguous accounting treatments, material estimates, or decisions requiring deep organizational and contextual knowledge. AI adoption should therefore be based on an assessment of task-AI fit, rather than on technological capability alone.

The accounting literature increasingly supports this task- and process-oriented perspective. Abbas (2026), in a comprehensive review of 91 studies, argues that digitalization and AI technologies are transforming management accounting not only through the automation of individual activities but also through changes in accounting information, organizational structures, relationships with other organizational functions, and professional roles. The significance of AI consequently extends beyond the efficiency of accounting procedures: it can alter how accounting information is generated, interpreted, communicated, and incorporated into managerial decision-making. Abbas further identifies new opportunities for multidisciplinary collaboration and emphasizes the emergence of new competencies and professional boundaries. This broader transformation means that AI adoption has implications for the design of the finance function itself rather than merely for individual activities within it.

The transformation of finance can therefore be understood as occurring across three increasingly consequential levels. The first is automation, where AI and related technologies perform predefined or repetitive activities such as invoice processing, transaction classification, reconciliation, and document extraction. The principal benefits at this level are typically reductions in processing costs, manual effort, cycle times, and errors. The second is augmentation, where AI supports human analysis and professional judgment through applications such as forecasting, anomaly detection, scenario analysis, management reporting, and financial interpretation. Here, the principal source of value is not labour substitution but improved productivity, analytical capacity, and potentially decision quality. The third is process orchestration, where AI capabilities are embedded across multiple stages of an end-to-end finance process. Emerging agentic systems illustrate this possibility by potentially coordinating information retrieval, analysis, exception identification, communication, and execution across several interconnected activities. At this level, the opportunity is not simply to make an existing process faster but to redesign the process itself.

This progression is consistent with Dong, Stratopoulos, and Wang (2024), whose review of research on ChatGPT and LLMs in accounting and finance identifies three interconnected streams: applications of LLMs across accounting and finance domains; the use of LLMs as information-processing and analytical tools; and the broader implications of LLM adoption for accounting and finance professionals, organizations, and sectors. The significance of this classification extends beyond the academic literature because it demonstrates that AI is simultaneously becoming a production technology, an analytical capability, and an organizational technology. The consequences for finance therefore extend from individual task execution to the structure of work and the allocation of responsibilities between humans and machines.

Stratopoulos and Wang (2025) develop this perspective further by emphasizing the implications of AI for accounting research and practice. Their framework highlights the expanding capabilities of generative AI and AI agents and the corresponding importance of human capabilities such as judgment, creativity, critical evaluation, and theoretical and contextual integration. This argument has direct implications for corporate finance. As AI becomes increasingly capable of producing information, conducting routine analysis, and generating preliminary outputs, the comparative value of finance professionals may shift toward activities that require contextual interpretation, critical challenge, professional judgment, and integration of financial information into broader business decisions.

This shift has important implications for how CFOs should evaluate AI investments. A narrow automation perspective tends to focus on the number of tasks that can be automated and the associated labour savings. An organizational-transformation perspective instead asks whether AI changes the economics and effectiveness of the end-to-end finance process. For example, the value of AI-enabled financial close should not be measured solely by the number of journal entries processed automatically. It should also encompass the speed of the close, the quality of reconciliations, the identification of exceptions, the time available for analysis, the quality of management information, and the resulting improvement in decision cycles. Similarly, the value of AI-enabled forecasting should not be reduced to statistical forecast accuracy. It should also be assessed according to whether managers receive relevant information earlier, evaluate a wider range of scenarios, and make better resource-allocation decisions.

This process perspective reinforces a central proposition of the CFO AI Navigator 2026. Parnova Consulting (2026) argues that AI should be treated as a strategic finance and leadership issue rather than as a conventional IT project. The guide emphasizes CFO ownership because AI affects not only technology infrastructure but also finance processes, data quality, management information, internal controls, organizational capabilities, and decision-making. Its recommendation that CFOs prioritize a limited number of high-value use cases, establish baseline measures, define success criteria, and design governance from the beginning is therefore consistent with the emerging academic understanding of AI as an organizational transformation rather than simply a new software category.

The emphasis on CFO ownership is particularly important because the finance function sits at the intersection of information, performance measurement, resource , risk, and organizational control. The CFO is consequently responsible not only for determining whether an AI application is technically feasible but also for assessing whether it creates economic value, whether its outputs are sufficiently reliable for the intended decision, what level of human oversight is required, and how responsibilities should be redistributed between people and technology. AI strategy therefore becomes inseparable from finance strategy.

The organizational implications are significant. As AI takes over an increasing proportion of routine information-processing activities, the boundaries between finance, information technology, data analytics, and business functions may become less distinct. Abbas (2026) identifies changes in the relationship between management accounting and other organizational functions as an important consequence of AI-enabled transformation. Finance may consequently evolve from being primarily a producer and controller of financial information toward becoming an orchestrator of data, analytical capabilities, decision support, and business insight. This transformation strengthens the case for multidisciplinary collaboration between finance professionals, data specialists, IT functions, risk specialists, and business leaders.

The changing role of finance professionals reinforces this point. If AI can increasingly perform information retrieval, classification, summarization, drafting, and elements of analytical processing, the relative value of human expertise shifts toward activities that depend on context, judgment, challenge, interpretation, and accountability. Stratopoulos and Wang (2025) make a closely related argument in the accounting research domain: as AI democratizes lower-level capabilities, human comparative advantage increasingly lies in higher-order activities requiring judgment, creativity, and integration. In corporate finance, this suggests that AI may ultimately strengthen rather than diminish the strategic role of finance professionals, provided that organizations redesign roles and processes so that employees can move from routine information production toward higher-value decision support.

This interpretation is also consistent with the productivity evidence from Brynjolfsson et al. (2025). Their findings suggest that AI can function as a mechanism for expertise diffusion, allowing less-experienced employees to achieve performance levels closer to those of more experienced colleagues. In a finance context, such effects could be particularly valuable in activities such as financial analysis, reporting, management commentary, and routine forecasting support. However, the findings should not be interpreted as evidence that professional expertise becomes less important. Rather, they suggest that the nature of expertise may change. Experienced finance professionals may increasingly spend less time producing first-order analysis and more time validating outputs, interpreting exceptions, exercising judgment, and addressing complex cases.

The same qualification follows from Dell'Acqua et al. (2025). Because AI capabilities are uneven across tasks, effective human-AI collaboration requires organizations to understand where AI performs reliably and where human intervention remains essential. This suggests that the appropriate design principle for finance should not be maximum automation, but optimal allocation of tasks between humans and AI. Routine and highly structured activities may be candidates for extensive automation; analytical activities may benefit from human-AI augmentation; and high-stakes or highly contextual decisions may require human leadership with AI operating primarily as an advisory capability.

This perspective also changes how AI maturity should be understood. A finance function should not be considered AI mature simply because it has deployed numerous AI tools or launched multiple pilots. Maturity is better reflected in its ability to integrate AI into processes, data architectures, management routines, control systems, and professional roles. AI maturity therefore represents a progression from isolated experimentation, through repeatable and measurable applications, toward an integrated organizational capability in which AI is embedded within the finance operating model.

The central proposition of the Parnova framework—that AI adoption should be owned by finance leadership—is therefore supported and sharpened by the academic literature. CFO ownership is not justified merely because finance happens to be an important user of AI. It is justified because AI changes the underlying configuration of the finance function: what information is produced, how it is processed, who interprets it, how decisions are made, and where accountability resides. The CFO's role is consequently to orchestrate the alignment of technological capability with financial objectives, organizational processes, professional expertise, and control requirements.

AI in finance should therefore be understood as a socio-technical transformation. Technology is a necessary component, but it is only one component. Sustainable value arises when AI capabilities are combined with high-quality data, redesigned processes, appropriate human expertise, effective governance, and clearly defined decision rights. The strategic unit of AI transformation is consequently not the algorithm, application, or individual task, but the end-to-end finance process in which technology, people, information, controls, and decisions interact.

Under this perspective, the CFO's task is not to maximize automation. It is to determine where automation, augmentation, and emerging AI-enabled orchestration can collectively produce superior finance-function performance while preserving professional judgment, organizational accountability, and control.

3. The Business Case for AI: From Hype to Measurable Value

The strategic potential of artificial intelligence (AI) in finance is increasingly well established, but the translation of technological potential into measurable economic value remains far from automatic. The CFO AI Navigator 2026 identifies a broad portfolio of finance applications, including invoice processing, financial close, expense and compliance management, forecasting and scenario analysis, margin management, liquidity forecasting, fraud and anomaly detection, conversational AI, reporting, and emerging agent-based process orchestration (Parnova Consulting, 2026). These applications differ substantially in their technological maturity, economic logic, implementation complexity, and risk profile. Consequently, a credible business case for AI should not ask simply whether a technology can perform a particular finance activity. It should establish where AI creates measurable incremental value relative to the existing process, under what organizational conditions, and at what level of risk.

This distinction is important because the current AI environment is characterized by considerable enthusiasm concerning productivity and transformation, while the empirical evidence remains highly context-dependent. Noy and Zhang (2023) demonstrate that generative AI can substantially improve productivity in professional knowledge work, while Brynjolfsson, Li, and Raymond (2025) show significant productivity gains in a large-scale field setting. At the same time, Dell'Acqua et al. (2025) demonstrate that AI performance is uneven across tasks, with benefits concentrated within the technology's capability frontier and potential performance deterioration outside it. The implication is that AI business cases should be constructed around specific tasks and processes, rather than extrapolated from generalized claims about AI productivity.

For CFOs, this suggests that the business case for AI can be understood through three increasingly sophisticated forms of value creation. The first concerns transactional automation, where AI reduces the cost and effort associated with repetitive activities. The second concerns analytical augmentation, where AI improves the speed, breadth, or quality of human analysis and decision-making. The third concerns generative and agentic finance, where AI may fundamentally restructure how finance processes are executed and coordinated. These categories are analytically distinct but not mutually exclusive. In practice, a mature AI-enabled finance function is likely to combine all three.

3.1 Transactional Automation

The most immediate business case for AI in finance arises in highly repetitive, structured, and relatively standardized processes. Accounts payable, invoice extraction, expense processing, transaction classification, reconciliation, and selected components of the financial close are natural candidates because they involve substantial transaction volumes, recurring procedures, and outputs that can often be evaluated against clearly defined rules or historical outcomes.

The economic logic of these applications is relatively straightforward. If AI reduces the amount of human effort required to process a transaction while maintaining or improving accuracy and control quality, the organization can potentially reduce processing costs, increase throughput, shorten cycle times, and redeploy employee capacity toward higher-value activities. For this reason, transactional automation is often an attractive entry point for finance AI because its benefits can be linked to observable operational measures such as processing cost per transaction, processing time, manual hours, exception rates, error rates, straight-through-processing rates, and control failures.

The CFO AI Navigator 2026 reports substantial potential efficiency gains in areas such as automated invoice processing and financial close, including reductions in processing costs and close-cycle duration (Parnova Consulting, 2026). Such figures are useful as practitioner-oriented indications of potential value, but they should not be interpreted as universal benchmarks. The realized economic return from an AI implementation is highly dependent on the organization's starting point. A company with highly manual accounts-payable processes may have considerable automation potential, whereas an organization that has already implemented sophisticated enterprise resource planning, workflow automation, and optical-character-recognition technologies may obtain only incremental benefits from an additional AI layer.

The business case should therefore be calculated against the actual baseline process, rather than against an abstract manual process. Relevant variables include transaction volumes, labour costs, existing automation, exception rates, system-integration requirements, implementation costs, data quality, vendor costs, and the time employees must spend validating AI outputs. This is particularly important because gross automation savings can overstate the true economic benefit if organizations fail to account for monitoring, exception management, human validation, and governance.

The broader productivity literature nevertheless provides evidence that AI can generate meaningful improvements in knowledge-intensive work. Noy and Zhang (2023) found that access to generative AI reduced average task completion time by approximately 40% while increasing output quality by approximately 18% in their experimental setting. These results provide evidence that generative AI can simultaneously affect both productivity and quality rather than merely reducing labour input. Brynjolfsson et al. (2025) similarly found an average productivity increase of approximately 15% among customer-support workers using generative AI, although the magnitude of the effect varied considerably across workers. These findings reinforce the proposition that AI can produce measurable economic benefits while simultaneously demonstrating why those benefits cannot be assumed to be constant across organizational contexts.

For finance leaders, the relevant question is therefore not whether AI can automate a particular activity. The more important question is:

Which finance processes contain sufficient volumes of repeatable work, sufficiently reliable data, and sufficiently measurable outcomes to justify AI-enabled redesign?

This question introduces an important shift from technology selection to economic process selection. The best initial AI use cases are not necessarily those involving the most advanced technology. They are those where the organization can clearly establish a baseline, identify the mechanism through which AI will create value, measure the resulting improvement, and control the associated risks.

The distinction between gross and net productivity is particularly important. An AI system may reduce the time required to produce an output while simultaneously increasing the time required for employees to verify that output. Consequently, the relevant measure is not simply time saved by the AI system but the net reduction in total process effort. CFOs should therefore assess the complete process, including the time required for validation, exception handling, monitoring, and governance. This approach provides a more realistic basis for evaluating whether an AI application generates genuine productivity improvements.

3.2 Analytical Augmentation

A second and potentially more strategically significant category consists of applications in which AI supports rather than replaces professional judgment. Forecasting, scenario analysis, liquidity management, anomaly detection, margin analysis, management reporting, and financial planning fall into this category. Here, the value proposition is fundamentally different from transactional automation. The objective is not primarily to eliminate human effort but to improve the organization’s ability to process information, identify patterns, evaluate alternatives, and make decisions (Abbas, 2026; Stratopoulos and Wang, 2025). The growing accounting literature similarly suggests that AI applications extend beyond automation to analytical and decision-support activities across management accounting and the wider accounting function (Dong, Stratopoulos and Wang, 2024; Abbas, 2026).

This distinction is particularly important for the finance function because many of its highest-value activities are characterized by uncertainty and managerial judgment. Forecasting, for example, is not simply a statistical exercise. A forecast must incorporate assumptions about demand, pricing, costs, market conditions, strategic initiatives, and management behaviour. AI can process larger quantities of historical and real-time information and generate alternative scenarios, but the resulting outputs still require interpretation and challenge (Abbas, 2026). This reflects the broader finding that AI capabilities can enhance information processing and analytical work while leaving important forms of professional and contextual judgment with human decision-makers (Stratopoulos and Wang, 2025).

Consequently, the business case for analytical AI cannot be evaluated solely through labour savings. Relevant performance measures should include forecast accuracy, forecast bias, planning-cycle duration, decision latency, scenario coverage, working-capital outcomes, loss avoidance, quality of management information, and the effectiveness of decisions made with AI support. The underlying question becomes whether AI improves the decision process, not merely whether it produces an analytically sophisticated output. This broader perspective is consistent with research showing that the organizational value of AI depends on how effectively its capabilities are incorporated into work processes rather than on technical performance alone (Noy and Zhang, 2023; Brynjolfsson, Li and Raymond, 2025; Dell’Acqua et al., 2026).

The distinction between model performance and organizational performance is critical. A statistically superior forecasting model does not automatically generate superior managerial decisions. Managers may misunderstand its assumptions, reject useful recommendations, over-trust erroneous outputs, or fail to act because the information arrives too late. AI-generated information therefore creates value only when it is embedded within an organizational process in which users can understand, challenge, and appropriately act upon the information. Research on AI in accounting emphasizes the importance of evaluating not only technical capabilities but also human interaction, organizational implementation, governance, and the conditions under which AI outputs can be appropriately used (Eisikovits, Johnson and Markelevich, 2024; Almeida and Santos Júnior, 2025; Stratopoulos and Wang, 2025).

The management-accounting literature highlights precisely these issues. Abbas (2026) identifies trust, accuracy, explainability, acceptability of AI-generated forecasts, and the appropriate balance between human and machine judgment as important unresolved issues in the adoption of AI in management accounting. AI can increase the volume and speed of available information, but this does not eliminate the need for professional judgment. Instead, it changes where judgment is exercised and potentially increases the importance of validating and contextualizing machine-generated outputs (Abbas, 2026). This is consistent with the broader accounting literature, which identifies human–AI interaction, explainability, professional responsibility, and the organizational consequences of AI as important areas for further research (Dong, Stratopoulos and Wang, 2024; Eisikovits, Johnson and Markelevich, 2024).

The empirical productivity literature provides a further reason for caution. Dell’Acqua et al. (2026) demonstrate that generative AI operates within a “jagged technological frontier”: knowledge workers benefited when tasks fell within AI’s effective capability range but could perform worse when tasks extended beyond that frontier. The finding has direct implications for finance forecasting and analysis. An AI model may be highly effective at identifying historical patterns while performing poorly when structural breaks, unprecedented market conditions, strategic changes, or unusual transactions invalidate those patterns (Dell’Acqua et al., 2026). The implication is that AI effectiveness is task-dependent rather than uniform, reinforcing the need for finance professionals to recognize the boundaries of model competence.

For CFOs, this suggests that analytical AI should be implemented through a human-AI decision architecture. AI should generate information, identify patterns, test scenarios, and provide recommendations, while finance professionals retain responsibility for interpretation, challenge, contextualization, and material decisions. The objective is therefore not to remove human judgment but to make human judgment more informed and efficient. Such an approach is consistent with the emerging governance literature, which emphasizes that AI adoption requires organizational structures, oversight mechanisms, accountability, and processes through which AI systems can be reviewed and challenged (Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025). It also reflects the accounting literature’s emphasis on balancing the capabilities of AI with the professional responsibilities and judgment of accountants (Abbas, 2026; Eisikovits, Johnson and Markelevich, 2024).

This principle is also important when assessing the distributional effects of AI within the finance workforce. Brynjolfsson, Li and Raymond (2025) find that less-experienced workers benefited particularly strongly from generative AI assistance, with AI increasing productivity on average while producing heterogeneous effects across workers. In finance, comparable effects could allow junior professionals to perform certain analytical and reporting activities more effectively, potentially reducing the time required to acquire routine expertise. At the same time, experienced professionals may increasingly shift toward reviewing outputs, handling exceptions, interpreting complex situations, and exercising judgment. AI can therefore alter not only productivity but also the structure and distribution of expertise within the finance function (Brynjolfsson, Li and Raymond, 2025; Abbas, 2026).

The broader productivity evidence reinforces the importance of considering both efficiency and quality when evaluating these workforce effects. Noy and Zhang (2023) provide experimental evidence that generative AI can increase productivity in professional writing tasks, while Brynjolfsson, Li and Raymond (2025) show that productivity effects vary according to worker experience and task characteristics. These findings suggest that the consequences of AI adoption should not be reduced to a simple assumption that automation produces uniform productivity gains. Instead, organizations need to examine how AI changes the interaction between technology, task characteristics, worker capabilities, and professional expertise (Noy and Zhang, 2023; Brynjolfsson, Li and Raymond, 2025; Dell’Acqua et al., 2026).

Analytical augmentation thus represents a transition from automation economics to decision economics. The value of AI is no longer primarily the labour that can be removed from a process, but the additional quality, speed, breadth, and consistency that can be introduced into managerial decision-making. This makes the measurement of AI value more complex but also potentially more strategically important. The emerging management-accounting literature supports this broader conception of AI value by emphasizing its implications for accounting information, organizational structures, professional roles, and business value rather than treating AI solely as a mechanism for reducing labour input (Abbas, 2026). Similarly, the wider accounting literature positions AI as both an analytical capability and a transformation of how accounting information is generated, interpreted, and used (Dong, Stratopoulos and Wang, 2024; Stratopoulos and Wang, 2025).

For CFOs and finance leaders, this implies that AI investment should ultimately be evaluated according to its contribution to organizational decision quality rather than technological sophistication alone. Governance, explainability, human oversight, validation, and clearly defined accountability mechanisms become integral components of the business case, particularly where AI outputs influence material financial or strategic decisions (Eisikovits, Johnson and Markelevich, 2024; Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025). Practitioner guidance similarly emphasizes the growing need for CFOs and boards to approach AI as a strategic capability requiring appropriate governance, implementation, and performance evaluation rather than as a stand-alone technology investment (Parnova Consulting, 2026).

Overall, the literature indicates that the strategic opportunity from analytical AI lies in augmenting managerial intelligence rather than simply automating managerial activity. AI can expand the volume of information available to finance professionals, accelerate analysis, identify patterns that may otherwise remain hidden, and support the exploration of alternative scenarios. However, these benefits are contingent on the ability of organizations to match AI capabilities with appropriate tasks, maintain human oversight, and establish governance mechanisms that prevent inappropriate reliance on machine-generated outputs (Abbas, 2026; Dell’Acqua et al., 2026; Almeida and Santos Júnior, 2025). The resulting model is therefore best understood as a human–AI system in which technological capability and professional judgment are complementary rather than competing sources of decision value.

3.3 Generative and Agentic Finance

The third category is potentially the most transformative because generative AI and emerging agentic systems can affect not only individual finance tasks but also the architecture of finance processes. Generative AI can interpret unstructured information, produce narrative explanations, interact with users, synthesize multiple information sources, and generate first-draft outputs. Agentic systems potentially extend these capabilities by coordinating multiple activities and interacting with enterprise systems to execute sequences of actions subject to specified constraints and human oversight.

The academic literature indicates that this transition is already becoming an important area of accounting and finance research. Dong, Stratopoulos, and Wang (2024) demonstrate that research on ChatGPT and LLMs in accounting and finance spans applications in financial reporting, auditing, taxation, accounting information systems, asset pricing, and other domains. Their review demonstrates that LLMs should not be viewed simply as advanced text-generation tools; they are increasingly being investigated as technologies that can influence how accounting and finance information is produced, processed, communicated, and used.

Stratopoulos and Wang (2025) similarly emphasize the expanding role of AI in accounting research and practice and point toward the increasing significance of AI agents and generative systems. Their analysis suggests that as AI capabilities become more sophisticated, the nature of accounting work and the competencies required of professionals will evolve. Domain expertise remains essential, but it increasingly needs to be combined with the ability to understand, evaluate, and effectively employ AI systems.

For finance functions, the implications could be substantial. Consider the financial close. Rather than using AI simply to automate one reconciliation or classify one transaction, an AI-enabled process could potentially retrieve supporting documentation, reconcile accounts, identify unusual movements, compare results with historical patterns, request explanations, draft reconciliation narratives, escalate unresolved exceptions, and prepare preliminary management reporting. The economic value in such a system would arise not from automating a single task but from compressing and restructuring the entire close process.

The same logic could apply to forecasting, treasury, tax, internal audit, and management reporting. An AI-enabled finance process might gather data from multiple systems, identify relevant developments, generate an analysis, construct scenarios, flag exceptions, and prepare an initial recommendation for human review. This represents a qualitative change in the nature of automation because the system is no longer merely performing a predefined task; it is participating in the coordination of a sequence of activities.

However, the potential of agentic finance should not be confused with organizational readiness. The more autonomous an AI system becomes, the greater the consequences of errors, inappropriate assumptions, unauthorized actions, or failures of control. A system that generates a draft management report creates a different risk profile from one that can alter a transaction, communicate with an external party, initiate a payment, or modify financial records.

Agentic AI therefore increases, rather than reduces, the importance of governance. Eisikovits, Johnson, and Markelevich (2024) emphasize that the adoption of AI in accounting and auditing creates important questions concerning reliability, accountability, privacy, professional judgment, and appropriate human oversight. These concerns become more significant as AI moves from generating recommendations toward executing actions. The fundamental governance question changes from "Can we trust this output?" to "Under what conditions should this system be permitted to act?"

This distinction has direct implications for the design of finance controls. Traditional controls are generally organized around human decision-makers and relatively predictable information systems. Agentic AI introduces a new control problem in which systems may dynamically determine which actions to take based on information, rules, model outputs, and interactions with other systems. Finance organizations will therefore need controls covering not only the accuracy of individual outputs but also system permissions, escalation rules, audit trails, action boundaries, monitoring, exception handling, and human approval thresholds.

Research on AI governance supports this broader interpretation. Almeida and Santos Júnior (2025) demonstrate that effective AI governance depends on organizational structures, processes, and capabilities rather than on technical mechanisms alone. Hadley, Blatecky, and Comfort (2025) similarly emphasize the importance of organizational support and the integration of responsible-AI governance into existing organizational processes. For finance, this implies that governance cannot be treated as a final compliance review after an AI system has been developed. Governance must be incorporated into the design of the process from the outset.

The business case for generative and agentic finance must therefore incorporate a wider range of benefits and costs than conventional automation. Potential benefits include reduced cycle times, lower manual effort, improved information availability, faster exception resolution, greater analytical capacity, and potentially new operating models. Potential costs include implementation complexity, integration requirements, model monitoring, human validation, cybersecurity, data protection, governance, and the financial consequences of erroneous autonomous actions.

The distinction between these benefits and costs is particularly important when comparing AI applications with different levels of autonomy. A generative AI system used to prepare a first draft of management commentary generally presents a different risk-return profile from an agentic system authorized to initiate transactions or alter financial records. As autonomy increases, the potential economic benefits may increase, but so too do the requirements for control, monitoring, accountability, and human intervention.

Across all three categories—transactional automation, analytical augmentation, and generative or agentic finance—the central managerial principle is therefore the same: AI investment should be evaluated according to realized process and organizational value rather than technological capability alone. The CFO AI Navigator 2026 makes a closely related argument by emphasizing focused use-case selection, baseline measurement, clear success criteria, and scalability rather than the proliferation of disconnected pilots (Parnova Consulting, 2026).

This perspective also provides a more rigorous interpretation of AI maturity. A finance function should not be considered mature because it has implemented the largest number of AI applications. Maturity is better reflected in its ability to identify high-value processes, measure baseline performance, select an appropriate degree of automation or augmentation, establish effective human oversight, and scale successful applications within an appropriate governance framework.

The business case for AI in finance is therefore ultimately a portfolio and process-design problem. Transactional automation can generate relatively immediate efficiency gains; analytical augmentation can improve the quality and speed of financial decisions; and generative and agentic technologies may enable deeper process transformation. The challenge for CFOs is to balance these opportunities according to organizational readiness, economic value, and risk. Moving from AI hype to measurable value therefore requires a fundamental shift in managerial thinking: the objective is not to maximize the deployment of AI, but to maximize the value created by appropriately governed AI-enabled finance processes.

4. The CFO as AI Transformation Owner

One of the strongest propositions of the CFO AI Navigator 2026 is that the CFO should own the transformation of the finance function through AI rather than delegate the initiative to the IT function (Parnova Consulting, 2026). This proposition should not be interpreted as a claim that CFOs need to become AI engineers or assume responsibility for technical implementation. Rather, it reflects a fundamental distinction between technology ownership and business accountability. IT and data functions provide essential technological capabilities, but finance leadership must determine how those capabilities should be translated into economic value, organizational change, and controlled decision-making.

This distinction becomes increasingly important as AI moves beyond conventional automation. IT functions are typically responsible for technology architecture, systems integration, cybersecurity, infrastructure, vendor management, access controls, and technical implementation. These responsibilities remain indispensable. However, they do not determine whether a particular finance process should be automated, augmented, or redesigned; whether an AI-generated recommendation is sufficiently reliable for a material financial decision; or whether the expected benefits justify the associated organizational and control risks. Those are fundamentally business and finance questions.

The CFO is uniquely positioned to address these questions because the finance function connects strategy, resource allocation, performance measurement, risk management, internal control, and management information. CFO leadership therefore involves determining which finance problems are economically significant, which processes offer meaningful opportunities for AI-enabled improvement, where professional judgment must remain central, what degree of error is acceptable, how AI-generated outputs should be validated, and how the resulting benefits should be measured. It also requires determining how responsibilities, skills, and roles within the finance organization should evolve as AI assumes a greater share of information-processing activities.

This makes AI adoption fundamentally different from a conventional technology investment. A traditional software implementation may primarily involve decisions about functionality, cost, integration, and technical performance. AI introduces additional questions concerning uncertainty, probabilistic outputs, model behaviour, explainability, accountability, and the appropriate distribution of decision authority between humans and machines. The CFO therefore has to evaluate not only whether an AI system works, but also whether it is appropriate for the decision context in which it will be used.

This issue becomes particularly important because AI may be technically capable of performing a task before an organization has established whether it should perform that task autonomously. A system may, for example, be capable of generating a management report, identifying unusual transactions, recommending accounting treatments, or initiating a workflow. Technical feasibility does not establish organizational legitimacy. The appropriate level of autonomy depends on factors such as financial materiality, regulatory requirements, reversibility, uncertainty, data sensitivity, and the consequences of an erroneous decision.

The accounting literature reinforces this concern. Eisikovits, Johnson, and Markelevich (2024) identify a range of risks and opportunities associated with AI adoption in accounting and auditing, including questions surrounding data ownership, governance, bias, professional trust, and potential deskilling. These issues demonstrate why AI implementation cannot be reduced to technical deployment. The organization must establish who is responsible for AI-generated outputs, who has authority to challenge them, how errors are identified and corrected, and where human expertise remains indispensable.

The challenge is particularly pronounced in finance because the consequences of errors can extend beyond operational inconvenience. An incorrect invoice classification may be relatively easy to detect and correct, whereas an inappropriate AI-supported forecast, liquidity recommendation, accounting judgment, or fraud assessment could affect materially significant decisions. The degree of human oversight should therefore be proportional to the materiality, uncertainty, and potential consequences of the AI-supported activity.

The CFO's role is consequently not to maximize the autonomy of AI but to determine the appropriate allocation of decision rights between humans and machines. In low-risk, highly standardized processes, substantial automation may be appropriate. In analytical processes involving uncertainty, AI may be more appropriately deployed as a decision-support capability, with finance professionals responsible for interpretation and approval. In high-impact decisions involving material estimates, significant financial consequences, or regulatory obligations, human accountability may need to remain decisive even where AI provides substantial analytical support.

This perspective also changes the meaning of CFO oversight. Effective leadership does not require the CFO to understand the technical details of every model or algorithm. It requires sufficient AI literacy to challenge assumptions, understand limitations, evaluate evidence of performance, and ask whether the technology is being used appropriately. The CFO must be able to distinguish between technical accuracy and business usefulness, between model performance and decision quality, and between productivity gains and genuine economic value.

The importance of this organizational capability is supported by the broader governance literature. Almeida and Santos Júnior (2025) show that AI governance is not simply a technical problem but depends on organizational structures, processes, responsibilities, and institutional capabilities. Their analysis reinforces the view that responsible AI requires governance arrangements that connect technological systems to organizational objectives and accountability. Similarly, Hadley, Blatecky, and Comfort (2025) examine the role of algorithm review boards in responsible AI governance and demonstrate the importance of organizational mechanisms capable of reviewing and challenging AI-related decisions. Although these studies are not specific to corporate finance, their implications are directly relevant to CFO-led AI governance.

For finance organizations, this suggests that AI governance should be embedded into existing management and control structures rather than treated as a separate technology exercise. The CFO, together with the CIO or CTO, risk management, legal, compliance, internal audit, and relevant business owners, should establish clear responsibilities for AI systems throughout their lifecycle. These responsibilities should extend from use-case selection and model development or procurement through validation, deployment, monitoring, modification, and eventual retirement.

The CFO AI Navigator 2026 is particularly relevant in emphasizing that governance should be designed early rather than added after successful experimentation (Parnova Consulting, 2026). This principle is important because governance requirements can affect the economic feasibility of an AI application from the beginning. Data-access restrictions, privacy requirements, model-validation procedures, human-review requirements, auditability, and integration constraints can materially alter implementation costs and expected benefits. Governance is therefore not merely a compliance cost; it is part of the business case itself.

CFO ownership also has implications for the management of AI investment as a portfolio. Rather than treating each AI initiative as an isolated technology project, finance leadership can evaluate investments across different horizons. Some applications may deliver relatively immediate productivity improvements through transactional automation. Others may require greater organizational change but offer improvements in forecasting, decision support, or control effectiveness. More transformative initiatives, including agentic process orchestration, may have greater long-term potential but also greater uncertainty and governance requirements. Portfolio-level management allows the CFO to balance short-term returns with longer-term strategic experimentation.

This portfolio perspective is consistent with the empirical evidence that AI benefits are heterogeneous. Noy and Zhang (2023), Brynjolfsson et al. (2025), and Dell'Acqua et al. (2025) all demonstrate, in different ways, that AI's effects depend on the characteristics of the task and the interaction between humans and technology. A CFO should therefore resist both excessive optimism and excessive skepticism. The relevant question is not whether AI "works" in general, but where it works, for whom, under what conditions, and with what consequences.

The CFO's responsibility consequently extends beyond selecting promising use cases. It encompasses the creation of an organizational environment in which AI can generate sustainable value. This includes establishing clear ownership, allocating resources, developing finance employees' AI-related capabilities, redesigning processes, defining performance measures, and ensuring that governance mechanisms evolve alongside technological capabilities. In this respect, AI transformation is as much a leadership and organizational-design challenge as it is a technology challenge.

The emerging literature on management accounting further supports this conclusion. Abbas (2026) emphasizes that AI is changing management accounting information, organizational structures, professional roles, and the competencies required of accounting professionals. Dong, Stratopoulos, and Wang (2024) similarly show that the implications of LLMs in accounting and finance extend beyond individual applications to broader questions concerning professionals and organizations. Stratopoulos and Wang (2025) further highlight the growing importance of combining accounting expertise with AI capabilities. Taken together, these studies suggest that AI transformation requires finance leaders to manage not only technology adoption but also the evolution of the finance profession itself.

This evolution creates an important leadership challenge. As AI becomes more capable of performing routine analytical and information-processing tasks, finance professionals may experience both productivity gains and changes in the nature of their work. Brynjolfsson et al. (2025) provide evidence that generative AI can particularly benefit less-experienced workers, potentially accelerating the diffusion of organizational knowledge. For CFOs, the strategic opportunity is therefore not simply to reduce headcount but to redesign work so that employees can devote more time to interpretation, business partnering, challenge, strategic analysis, and decision support. Conversely, failure to redesign roles may result in employees becoming overly dependent on AI outputs or gradually losing important domain expertise.

This concern is particularly relevant to the risk of deskilling identified by Eisikovits et al. (2024). If AI systems routinely perform analytical and judgment-related tasks, employees may have fewer opportunities to develop the underlying expertise required to challenge those systems effectively. The organization could therefore face a paradox in which greater AI capability reduces the human expertise needed to supervise it. CFO leadership must address this risk through deliberate capability development, ensuring that employees understand both how to use AI and how to critically evaluate its outputs.

The appropriate objective is therefore human-AI complementarity rather than simple substitution. AI should expand the analytical capacity of finance while preserving the professional capabilities necessary to interpret, challenge, and govern its outputs. This requires a deliberate allocation of tasks and decision rights based on the relative strengths of humans and AI. AI is generally well suited to processing large volumes of information, identifying patterns, generating preliminary outputs, and supporting repetitive analytical activities. Humans remain essential where contextual knowledge, accountability, ethical judgment, ambiguity, and organizational consequences are central.

The CFO's role can therefore be conceptualized as the design and stewardship of a decision architecture around AI. This architecture determines which processes AI can execute autonomously, which require human review, which decisions require explicit human approval, how exceptions are escalated, how AI performance is monitored, and who remains accountable for the final outcome. The focus shifts from managing individual AI tools to governing the interaction between technology, people, processes, and decisions.

This is ultimately the strongest interpretation of the Parnova proposition. CFO ownership does not mean replacing the expertise of IT, data science, risk, or compliance functions. Instead, it means ensuring that these capabilities are coordinated around finance's strategic objectives. The CFO provides the economic and organizational logic within which technical capabilities are deployed, while specialist functions provide the expertise required to implement and control them.

The central implication is therefore that AI transformation should be business-led, technically enabled, and jointly governed. The CFO should own the strategic outcomes and decision architecture; technology functions should own technical architecture and implementation; risk, legal, compliance, and internal audit should provide appropriate independent challenge; and finance professionals should retain accountability for material judgments and decisions. Such a model avoids both extremes: treating AI as an IT project on the one hand and expecting finance leadership to manage technical systems without specialist support on the other.

The CFO's ultimate responsibility is thus not to select the most sophisticated AI technology. It is to determine where AI should be used, how it should be used, where it should not be used, and how the organization can capture its benefits without compromising financial control, professional judgment, or accountability. In this sense, the CFO becomes not the technical owner of AI, but the strategic owner of AI-enabled finance transformation.

5. Why Fewer, Better Use Cases May Outperform Broad AI Experimentation

One of the recurring implementation risks identified by the CFO AI Navigator 2026 is the proliferation of disconnected AI pilots. This concern is particularly important because finance departments can typically identify a large number of potential AI applications. Invoice processing, financial close, forecasting, reporting, anomaly detection, treasury, tax, internal audit, and management support all provide plausible opportunities for AI deployment. The existence of numerous potential use cases, however, does not imply that organizations should pursue them simultaneously.

The proliferation of pilots creates a significant organizational coordination problem. Each additional AI initiative can require access to data, systems integration, cybersecurity assessment, vendor evaluation, model validation, user training, governance arrangements, and change management. As the number of initiatives increases, these requirements can fragment managerial attention and stretch organizational capabilities. An organization may therefore appear highly active in AI while producing relatively little measurable business value.

The relevant objective is consequently not to maximize the number of AI experiments. It is to maximize the number of scalable AI-enabled processes that produce measurable and sustainable business value. This distinction between experimentation and value creation is central to the Parnova framework, which emphasizes focused use-case selection, baseline measurement, clear success criteria, and scalability rather than broad but disconnected experimentation (Parnova Consulting, 2026).

The academic literature provides support for this more selective approach. The empirical evidence on generative AI demonstrates that performance benefits are highly contingent on the characteristics of the task and the organizational context in which the technology is used. Noy and Zhang (2023) demonstrate substantial productivity improvements in specific knowledge-work tasks, while Brynjolfsson et al. (2025) find significant but heterogeneous productivity gains across workers. Dell'Acqua et al. (2025) further demonstrate that AI performance varies substantially depending on whether a task falls within the technology's effective capability frontier. Taken together, these findings suggest that organizations should not assume that every plausible AI application will produce equivalent benefits.

Prioritization should therefore begin with the characteristics of the finance problem, rather than the availability or novelty of the technology. This problem-oriented approach is consistent with the emerging AI-in-accounting literature, which emphasizes the importance of aligning AI applications with organizational objectives, task characteristics, implementation conditions, and professional requirements rather than treating AI adoption as a technology-led exercise (Dong, Stratopoulos and Wang, 2024; Stratopoulos and Wang, 2025). A useful use-case assessment should consider four interrelated dimensions: expected economic impact, organizational and technical feasibility, strategic relevance, and risk and transformation complexity.

Economic impact should encompass more than direct labour savings. Some AI applications generate value primarily through reduced processing costs or shorter cycle times, while others may improve decision quality, reduce financial losses, strengthen control effectiveness, or increase the capacity of finance professionals to perform higher-value activities. Research on generative AI demonstrates that productivity improvements can arise through increased task efficiency, but the magnitude and nature of these benefits vary according to the task and the worker (Noy and Zhang, 2023; Brynjolfsson, Li and Raymond, 2025). Similarly, evidence from accounting research suggests that AI creates value through multiple channels, including automation, information processing, analytical support, and decision assistance (Dong, Stratopoulos and Wang, 2024; Abbas, 2026). For this reason, CFOs should distinguish between applications that primarily generate efficiency gains and those that create value through improved financial decisions.

Feasibility concerns the organization’s capacity to implement and operate an AI application effectively. Relevant considerations include the availability and quality of data, the maturity and standardization of the underlying process, system-integration requirements, technical capabilities, and the availability of employees who can manage and use the resulting system. AI implementation is therefore not solely a question of whether a model can technically perform a task; organizational readiness, governance arrangements, data conditions, and user capabilities can materially influence whether the technology generates value in practice (Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025). A use case with high theoretical value but poor data quality or extensive integration requirements may be less attractive than a more modest application that can be implemented reliably and scaled across the organization. This is particularly important because the performance of AI systems can vary substantially according to task conditions and the boundaries of their effective capabilities (Dell’Acqua et al., 2026).

Strategic relevance captures the relationship between an AI initiative and the broader priorities of the finance function and the organization. A highly innovative application may have limited value if it addresses a marginal activity, whereas a relatively simple AI application may be strategically important if it improves a core process such as financial close, forecasting, working-capital management, or management reporting. The accounting literature increasingly frames AI as a transformation of accounting information processes and organizational decision-making rather than merely as a collection of isolated technological applications (Abbas, 2026; Stratopoulos and Wang, 2025). AI investment should therefore be evaluated as part of the finance strategy rather than as an independent technology agenda. This also implies that use-case selection should reflect the organization’s strategic priorities, operating model, and intended role for the finance function.

Risk and transformation complexity represent a fourth critical dimension. Risk can include operational, regulatory, privacy, cybersecurity, model, and reputational exposure. Transformation complexity encompasses the extent of process redesign, systems integration, organizational change, training, and governance required to move an application from pilot to routine operation. Research on AI governance emphasizes that organizations require appropriate structures for accountability, oversight, review, and risk management as AI becomes embedded in organizational processes (Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025). Accounting research likewise identifies concerns relating to accuracy, explainability, professional responsibility, and the appropriate balance between human and machine judgment (Eisikovits, Johnson and Markelevich, 2024; Abbas, 2026). A technologically impressive use case may therefore be inappropriate as an early investment if it requires extensive organizational change and creates substantial control challenges.

This prioritization logic leads naturally to a portfolio approach rather than the management of AI as a single project. A finance AI portfolio can contain several categories of investment with different time horizons, risk profiles, and sources of value. Such a portfolio perspective is consistent with the broader accounting research agenda, which identifies AI applications ranging from task-level automation and decision support to more fundamental changes in accounting processes and organizational structures (Dong, Stratopoulos and Wang, 2024; Stratopoulos and Wang, 2025).

The first category consists of productivity-oriented applications. These include use cases such as invoice processing, document analysis, routine reporting support, and other applications where AI can reduce manual effort or accelerate information processing. Their principal value lies in measurable improvements in time, cost, throughput, and employee productivity. Experimental evidence demonstrates that generative AI can produce substantial productivity improvements on appropriate knowledge-work tasks (Noy and Zhang, 2023), while field evidence indicates that these effects can be particularly beneficial for less-experienced workers (Brynjolfsson, Li and Raymond, 2025). Such applications are often appropriate early candidates because their benefits can be assessed against relatively clear operational baselines and because they can provide organizations with practical experience in AI adoption.

A second category consists of process-transformation applications. These include areas such as financial close, forecasting, liquidity management, expense management, and other processes in which AI may improve not only individual tasks but also the overall structure and speed of the process. The principal objective in this category is to redesign core finance activities in ways that improve cycle times, information quality, analytical capacity, and decision support. The management-accounting literature suggests that AI can increasingly influence how information is generated, analysed, and used in management processes, thereby extending its impact beyond isolated productivity improvements (Abbas, 2026). These initiatives generally require more substantial organizational change than productivity applications but may create more significant and sustainable value because they address end-to-end processes rather than individual activities.

A third category consists of strategic or transformative applications. Emerging agentic finance systems, autonomous exception handling, advanced scenario-generation capabilities, and highly integrated AI-enabled decision processes fall into this category. Their potential value lies in more fundamental changes to the finance operating model. However, these applications are also likely to involve greater uncertainty, governance requirements, integration complexity, and control risks. The governance literature indicates that increasing organizational reliance on AI requires corresponding mechanisms for oversight, accountability, and responsible implementation (Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025). Moreover, evidence concerning the “jagged technological frontier” suggests that organizations should be cautious about extending AI into tasks where its capabilities are uncertain or where errors may have significant consequences (Dell’Acqua et al., 2026). They should therefore be treated as selective long-term investments rather than as immediate replacements for core finance processes.

The value of this portfolio logic lies in its ability to balance exploitation and exploration. Productivity-oriented applications can generate relatively rapid and measurable benefits while allowing the organization to develop experience with AI technologies, data governance, user adoption, and performance measurement. Process-transformation initiatives can then build on these capabilities to redesign strategically important finance processes. More transformative applications can be pursued selectively as the organization’s technical and governance capabilities mature. This staged approach is consistent with evidence that AI value depends not simply on technological capability but on the interaction between technology, tasks, workers, organizational processes, and governance (Brynjolfsson, Li and Raymond, 2025; Dell’Acqua et al., 2026; Abbas, 2026).

Consequently, the portfolio should not be understood as a fixed sequence in which every organization must progress from automation to transformation at the same speed. Rather, it provides a framework for allocating scarce investment and management attention according to value, feasibility, strategic importance, and risk. A CFO can use this framework to ensure that early AI investments generate demonstrable organizational learning and economic value while creating the data, skills, governance capabilities, and organizational confidence required for more ambitious applications. In this sense, AI prioritization becomes an exercise in managing a portfolio of organizational capabilities, rather than simply selecting individual technologies (Stratopoulos and Wang, 2025; Almeida and Santos Júnior, 2025; Parnova Consulting, 2026).

This sequencing is particularly important because organizational learning is itself an important outcome of AI adoption. Abbas (2026) emphasizes that AI affects not only accounting tasks but also organizational structures, information flows, professional roles, and required competencies. Similarly, Dong, Stratopoulos, and Wang (2024) demonstrate that the adoption of LLMs has implications extending beyond individual applications to the broader organization and the accounting profession. An AI portfolio should therefore be evaluated not only according to the direct benefits of individual projects but also according to the capabilities that successful implementation develops for subsequent initiatives.

The portfolio approach also helps address the governance challenges associated with scaling AI. Almeida and Santos Júnior (2025) emphasize that AI governance depends on organizational structures, processes, and institutional capabilities. Hadley et al. (2025) similarly highlight the importance of organizational mechanisms for responsible AI oversight. These insights suggest that each new AI initiative should not require the organization to create an entirely new governance structure. Instead, organizations should progressively develop reusable capabilities for risk assessment, model review, data governance, monitoring, accountability, and human oversight.

This principle is especially relevant to the transition from pilot projects to scaled operations. A pilot may be technically successful while still being organizationally unsuitable for widespread deployment. An application that performs well with a small group of users may encounter new problems when it is exposed to larger data volumes, more diverse transactions, additional business units, or material financial decisions. Scaling therefore requires more than replicating a successful technical experiment. It requires demonstrating that the underlying process, controls, data architecture, governance arrangements, and user capabilities can operate reliably at organizational scale.

The distinction between pilot success and scalable value should consequently be central to CFO decision-making. A pilot should be designed from the outset with clear criteria concerning not only whether the technology performs as expected, but also whether it produces measurable business value, can be integrated into existing processes, meets governance requirements, and can realistically be expanded beyond the initial use case. This principle is consistent with the Parnova framework's emphasis on designing pilots for scalability rather than treating experimentation as an end in itself (Parnova Consulting, 2026).

The evidence concerning the heterogeneous effects of AI reinforces this selective approach. Brynjolfsson et al. (2025) demonstrate that AI productivity gains vary substantially between workers, while Dell'Acqua et al. (2025) show that technological effectiveness depends on the characteristics of the task. A portfolio approach therefore enables organizations to learn systematically about where AI performs effectively within their own finance environment. Rather than making a large, irreversible commitment based on generalized claims about AI, the organization can progressively identify which processes generate demonstrable value and which do not.

This approach also provides a mechanism for managing uncertainty. Not all AI initiatives will succeed, and not every promising application should be scaled. The appropriate response is not to avoid experimentation but to structure experimentation around measurable learning. Each initiative should clarify the baseline problem, expected value mechanism, relevant risks, success criteria, and potential path to scale. Applications that demonstrate measurable value and organizational fit can be expanded; those that do not can be modified or discontinued before substantial resources are committed.

The portfolio perspective is therefore preferable to both an all-or-nothing AI strategy and an uncontrolled proliferation of pilots. An all-or-nothing approach creates excessive dependence on a small number of uncertain technologies, while uncontrolled experimentation can dissipate resources across disconnected initiatives. A balanced portfolio allows organizations to capture relatively low-risk productivity gains while progressively developing the technological, organizational, and governance capabilities required for more consequential forms of AI-enabled transformation.

For CFOs, the resulting management principle is clear: AI investments should be prioritized according to business value, feasibility, strategic relevance, and risk—not according to technological novelty or the number of potential applications. The most effective AI agenda is therefore likely to be selective rather than expansive. Its success should be measured not by the number of pilots launched, but by the number of finance processes that have been demonstrably improved, responsibly governed, and successfully scaled across the organization.

6. Baselines and the Measurement Problem

A major strength of the CFO AI Navigator 2026 is its insistence that organizations establish baseline measurements before implementing AI (Parnova Consulting, 2026). This recommendation is particularly important because AI adoption can easily generate misleading perceptions of success. A finance team may conclude that an AI application “saves time” without establishing how much time was actually saved, whether the quality of the output changed, whether errors increased or decreased, or whether employees simply shifted their effort from producing outputs to validating and correcting AI-generated results.

The measurement problem is therefore more fundamental than simply selecting appropriate key performance indicators. It concerns the distinction between apparent activity, technical performance, and realized organizational value. An AI system may appear successful because it produces outputs rapidly, yet create limited economic value if substantial human effort is subsequently required to verify, correct, or contextualize those outputs. Conversely, an AI system may not generate dramatic reductions in labour hours but may nevertheless create significant value by improving the quality, consistency, or timeliness of financial decisions.

For this reason, the measurement of AI performance should begin with a clear understanding of the pre-AI process. Before implementation, finance leaders should establish relevant baselines concerning processing time, labour input, error rates, exception rates, cycle times, output quality, control performance, and, where relevant, decision outcomes. The appropriate baseline will depend on the use case. Invoice automation may require operational measures relating to processing costs and straight-through processing, whereas an AI-supported forecasting application may require measures relating to forecast accuracy, bias, planning speed, scenario coverage, and the quality of subsequent decisions.

The importance of such baseline measurement is reinforced by recent empirical research. Noy and Zhang (2023) provide evidence that generative AI can substantially improve productivity and output quality in professional knowledge work. Their findings demonstrate the potential for meaningful performance gains but should not be interpreted as evidence that the same effects will occur uniformly across all tasks and organizational settings. The value of AI depends on the interaction between the technology, the nature of the task, and the way in which users incorporate AI into their work.

Brynjolfsson, Li, and Raymond (2025) provide further evidence of this heterogeneity. Their field study shows that generative AI can generate substantial productivity improvements, but the magnitude of those gains differs across workers. In particular, less-experienced employees benefited disproportionately in their research setting. This finding is highly relevant for finance measurement because average productivity improvements may conceal significant variation across roles, levels of expertise, and activities. An AI application that produces substantial benefits for junior analysts may have limited benefits for highly experienced professionals, while another application may generate value primarily by enabling experienced employees to devote more time to complex analytical and strategic activities.

Dell'Acqua et al. (2025) provide an even stronger reason for rigorous measurement through their concept of the jagged technological frontier. Their field experiment involving knowledge workers demonstrates that AI can improve performance on tasks that fall within its effective capability frontier while potentially worsening performance on tasks outside that frontier. The implication is that organizations cannot assume that AI deployment will generate uniformly positive effects simply because the technology performs well in demonstrations or pilots. Performance must be measured against the specific tasks and decisions that employees actually undertake.

This finding has direct relevance for finance. An AI system may perform effectively when summarizing standardized financial information or identifying familiar patterns but less effectively when dealing with unusual transactions, ambiguous accounting treatments, structural breaks in forecasting, or highly contextual business decisions. Aggregated performance indicators may therefore obscure significant weaknesses in specific parts of the process. CFOs should consequently examine not only average performance but also the distribution of outcomes, the types of exceptions produced, and the circumstances under which the AI system performs poorly.

A KPI such as “hours saved” is therefore insufficient on its own. Gross reductions in the time required to complete a task do not necessarily represent genuine productivity gains. Employees may need to spend additional time reviewing AI outputs, correcting errors, investigating exceptions, documenting decisions, or complying with new governance requirements. The relevant measurement question is therefore whether the AI application reduces the total effort required to complete the process at an acceptable level of quality and control.

This distinction is especially important because AI can shift work rather than eliminate it. For example, an AI system may rapidly generate a draft management report that previously required several hours of manual preparation. However, if finance professionals subsequently need to spend considerable time checking the factual accuracy of the report, correcting inappropriate interpretations, and verifying the underlying data, the net productivity benefit may be substantially smaller than the apparent time savings suggest. In some circumstances, the redistribution of work may nevertheless be beneficial if validation requires more valuable professional judgment than routine drafting. The point is that the value cannot be inferred from the AI system's speed alone.

The same principle applies to forecasting and other analytical applications. An AI model may demonstrate improved statistical accuracy without necessarily improving organizational decision-making. Finance professionals and managers may fail to understand the assumptions underlying the model, over-rely on AI-generated forecasts, or receive the information too late for it to influence decisions. Alternatively, a modest improvement in forecast accuracy may create significant organizational value if it enables earlier intervention, improves working-capital management, or supports more effective resource allocation.

The measurement of analytical AI should therefore distinguish between model performance and decision performance. Model-level measures, such as accuracy or prediction error, remain important because they indicate whether the underlying system performs technically as intended. However, they do not establish whether the organization benefits economically from the model. A broader assessment should examine whether the AI-enabled process improves decision speed, scenario analysis, managerial understanding, resource allocation, financial outcomes, or risk management.

This distinction is consistent with the management-accounting literature. Abbas (2026) identifies questions relating to trust, accuracy, explainability, and the interaction between human and machine judgment as important areas for further research. These issues indicate that the performance of an AI system cannot be assessed independently of the organizational context in which its outputs are used. An accurate AI-generated forecast that managers do not trust may have little practical value. Equally, a trusted but poorly understood AI system may create risks if decision-makers become excessively dependent on its outputs.

Measurement should therefore capture both objective performance and organizational adoption. An AI application must not only produce reliable outputs; those outputs must be incorporated appropriately into finance processes and decision-making. Relevant indicators may therefore include the frequency with which employees override AI recommendations, the nature and causes of overrides, the types of errors detected during human review, user confidence, the time required to resolve exceptions, and the degree to which AI-generated insights influence actual decisions.

The importance of this approach becomes even greater as AI systems become more autonomous. A generative AI system used to prepare an initial draft can be evaluated primarily according to productivity and output quality. An AI system that recommends financial actions requires additional measures relating to the quality of recommendations and the consequences of decisions. An agentic system capable of executing multiple actions requires broader measurement relating to control effectiveness, exception management, compliance, escalation, and the consequences of autonomous behaviour.

This suggests that AI measurement should be aligned with the level of organizational autonomy granted to the technology. As AI moves from assisting employees toward influencing or executing decisions, the measurement framework must move beyond productivity indicators toward broader assessments of risk, control, accountability, and organizational outcomes.

The Parnova framework's emphasis on baseline measurement is therefore more significant than it may initially appear. Establishing a baseline creates the conditions necessary for distinguishing genuine improvement from technological enthusiasm. Without a baseline, organizations cannot determine whether AI has reduced process costs, improved quality, accelerated decision-making, or merely changed the way work is performed (Parnova Consulting, 2026).

A rigorous baseline should also capture more than the performance of the process immediately before implementation. Finance leaders should identify the relevant sources of variation in the process, including differences between business units, transaction types, employee groups, and levels of complexity. This is particularly important in light of the heterogeneous effects identified by Brynjolfsson et al. (2025) and the task-dependent performance identified by Dell'Acqua et al. (2025). An AI application may perform well on routine cases while generating substantial difficulties in exceptional cases. Average performance measures alone may therefore conceal operational or control risks.

The evaluation of AI should consequently be an ongoing process rather than a one-time comparison between conditions before and immediately after implementation. AI systems, data environments, user behaviour, and organizational processes can all change over time. A successful pilot may perform differently after being scaled across additional business units or exposed to more complex data. Continuous monitoring is therefore required to determine whether early performance gains persist, improve, or deteriorate as the application becomes embedded within the finance function.

The governance literature provides additional support for this approach. Almeida and Santos Júnior (2025) emphasize that AI governance requires organizational capabilities and processes capable of managing AI throughout its lifecycle. Hadley, Blatecky, and Comfort (2025) similarly highlight the need for organizational mechanisms that provide structured oversight of responsible AI. For CFOs, this means that performance measurement and governance should be integrated. The same information used to assess economic value should also help identify emerging risks, performance deterioration, unexpected errors, and inappropriate patterns of use.

This integrated approach has an important managerial implication: an AI project should not be declared successful merely because it is technically functional or because users report positive experiences. Success requires evidence that the application produces measurable improvements relative to a clearly defined baseline while maintaining or improving appropriate standards of quality, control, and accountability.

The most effective measurement frameworks therefore connect three levels of performance. The first is technical performance, concerning whether the AI system performs its intended function reliably. The second is process performance, concerning whether the finance process becomes faster, more efficient, more accurate, or more effective. The third is organizational and economic value, concerning whether improvements in the process ultimately contribute to better decisions, financial outcomes, risk management, or strategic capacity.

The distinction between these levels is central to avoiding one of the most common errors in AI implementation: confusing technological capability with organizational value. An AI system can be technically impressive without improving a finance process, and a more accurate model does not necessarily generate better decisions. Conversely, a relatively simple AI application can create substantial value if it addresses a clearly defined bottleneck in an important finance process.

The central principle is therefore that AI value must be measured against a baseline and evaluated at the level of the process and organization, not merely at the level of the model or tool. This principle strengthens the broader argument of the CFO AI Navigator 2026: AI implementation should be governed by clear business objectives, measurable success criteria, and sustained performance monitoring rather than by generalized assumptions about technological productivity.

In this sense, baseline measurement performs two essential functions. First, it provides an evidence-based method for determining whether an AI application has actually created value. Second, it creates organizational discipline by forcing managers to specify, before implementation, what problem the AI system is expected to solve and how success will be recognized. This discipline is particularly important in an environment characterized by considerable technological enthusiasm and uncertain performance outcomes.

Ultimately, the measurement problem leads to a broader conclusion. The success of AI in finance should not be measured by how frequently the technology is used, how sophisticated the underlying model is, or how many hours it appears to save. It should be measured by whether AI produces sustained improvements in finance-process performance, decision quality, and organizational value while maintaining appropriate standards of control, accountability, and professional judgment.

7. Human-AI Collaboration Rather Than Simple Substitution

The economic and organizational consequences of artificial intelligence cannot be adequately understood as a binary choice between automation and job replacement. Although AI can automate particular tasks, the available evidence suggests that its most significant effects frequently arise through the interaction between human capabilities and machine capabilities. The central managerial question is therefore not simply whether AI can replace human work, but how work should be reorganized when humans and AI systems perform different but interdependent roles.

This distinction is particularly important for finance. Finance work combines highly standardized activities, such as transaction processing and reconciliation, with activities requiring contextual interpretation, professional judgment, accountability, and communication. AI may be highly effective at processing information, identifying patterns, generating preliminary analyses, and supporting repetitive activities. However, the appropriate interpretation and use of those outputs frequently remain dependent on human expertise.

The empirical literature supports this more differentiated view of AI adoption. Brynjolfsson, Li, and Raymond (2025), in Generative AI at Work, find that generative AI can produce significant productivity gains but that those gains are distributed unevenly across workers. In their research setting, less-experienced employees benefited particularly strongly from AI assistance, suggesting that generative AI can provide access to forms of knowledge and guidance that would otherwise require greater experience. The study therefore indicates that AI may alter the distribution of expertise within organizations rather than simply replacing workers.

This finding has important implications for finance functions. Less-experienced professionals may use AI to accelerate routine analytical work, obtain explanations of complex information, prepare preliminary reports, or structure initial analyses. More experienced professionals, by contrast, may obtain relatively smaller productivity gains from activities that they already perform efficiently. Their role may increasingly shift toward validating outputs, resolving complex exceptions, exercising professional judgment, and interpreting information within its broader organizational context.

The relationship between AI and expertise is, however, more complex than a simple distinction between junior and senior employees. Dell'Acqua et al. (2025) demonstrate that AI performance depends substantially on the nature of the task. Their research on the "jagged technological frontier" shows that AI can improve performance when tasks fall within its effective capability range while potentially reducing performance when users rely on AI outside that range. Human expertise therefore remains essential not only for performing tasks that AI cannot reliably perform but also for determining when AI should and should not be relied upon.

For finance organizations, this points toward a human-in-the-loop operating model. Under such an approach, AI and finance professionals perform complementary roles. AI may collect, classify, and synthesize information; identify anomalies; produce preliminary forecasts; draft explanations; generate management reports; propose reconciliations; and identify potential control exceptions. Finance professionals, in turn, may validate material conclusions, interpret business context, challenge assumptions, assess uncertainty, determine appropriate accounting treatment, approve significant decisions, and communicate implications to management and boards. This complementary model is consistent with the accounting literature, which increasingly emphasizes that AI should augment rather than automatically replace professional judgment, particularly where accounting and financial decisions involve uncertainty and contextual interpretation (Abbas, 2026; Eisikovits, Johnson and Markelevich, 2024; Stratopoulos and Wang, 2025).

The purpose of this division is not to preserve human involvement for its own sake. Rather, it is to allocate work according to the comparative strengths of humans and AI systems. AI can process large volumes of information at speed and identify patterns that may be difficult for individuals to detect consistently, while human professionals contribute contextual understanding, organizational knowledge, ethical reasoning, accountability, and the ability to exercise judgment in ambiguous or unprecedented circumstances (Noy and Zhang, 2023; Brynjolfsson, Li and Raymond, 2025; Dell’Acqua et al., 2026). Evidence from the productivity literature indicates that generative AI can improve performance on tasks within its effective capability range, but its benefits are not uniform across all activities (Noy and Zhang, 2023; Dell’Acqua et al., 2026). The appropriate allocation of responsibilities must therefore reflect the characteristics and risks of individual finance tasks.

This allocation is particularly important in high-stakes financial decisions. An AI system may, for example, identify a potentially unusual transaction or recommend a particular forecasting scenario. However, determining whether the anomaly represents fraud, a legitimate business exception, or a data-quality problem may require knowledge of the organization’s operations and relationships. Similarly, an AI-generated forecast may identify a statistically plausible outcome without understanding a pending strategic decision, a planned acquisition, a change in management behaviour, or another event not adequately represented in historical data. These limitations reinforce the importance of professional challenge and contextual validation when AI outputs inform accounting or financial decisions (Abbas, 2026; Eisikovits, Johnson and Markelevich, 2024). They also reflect the “jagged technological frontier” identified by Dell’Acqua et al. (2026), whereby AI can generate substantial benefits on some tasks but may reduce performance when applied beyond the boundaries of its effective capabilities.

The governance implications are equally important. A human-in-the-loop model requires more than simply placing an employee at the end of an automated workflow. Organizations need clearly defined decision rights, escalation procedures, review thresholds, documentation requirements, and accountability for AI-assisted decisions. Research on AI governance emphasizes the importance of organizational structures and oversight mechanisms that enable AI systems to be monitored, reviewed, and appropriately challenged (Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025). Within finance, these mechanisms are particularly relevant because AI outputs may influence financial reporting, forecasts, controls, resource allocation, and other decisions with material organizational consequences (Eisikovits, Johnson and Markelevich, 2024; Abbas, 2026).

The appropriate objective is therefore not maximum automation but the optimal allocation of human and machine capabilities. This principle requires finance leaders to distinguish carefully between activities that can appropriately be automated, activities that should be augmented by AI, and activities in which human accountability and judgment must remain central. Such a distinction is consistent with the emerging view of AI in accounting as a transformation of the relationship between technology, professional expertise, and organizational decision-making rather than simply a replacement of human labour (Dong, Stratopoulos and Wang, 2024; Stratopoulos and Wang, 2025).

In practical terms, this suggests a three-level allocation of finance activities. Automation is most appropriate where tasks are repetitive, rules-based, highly standardized, and subject to readily observable quality controls. Augmentation is appropriate where AI can substantially improve information processing or analysis while human professionals remain responsible for interpretation and validation. Human-led decision-making should remain dominant where activities involve material financial consequences, significant uncertainty, ethical considerations, novel circumstances, or complex organizational context. This approach allows finance organizations to capture AI-driven productivity benefits without assuming that technological capability automatically translates into sound financial judgment (Abbas, 2026; Eisikovits, Johnson and Markelevich, 2024; Dell’Acqua et al., 2026).

Ultimately, the human-in-the-loop model should be understood as an operating-model choice rather than merely a control mechanism. Its objective is to ensure that AI performs the activities for which it has a comparative advantage while finance professionals retain responsibility for the areas in which human expertise provides greater value. As AI becomes more deeply embedded in finance processes, the role of finance professionals may consequently shift away from routine information processing toward validation, exception management, interpretation, communication, and judgment. The strategic challenge for CFOs is therefore to design workflows in which these human capabilities complement AI rather than simply acting as a final safeguard against it (Brynjolfsson, Li and Raymond, 2025; Abbas, 2026; Stratopoulos and Wang, 2025).

The management-accounting literature supports this conclusion. Abbas (2026) argues that AI is transforming accounting work, professional roles, organizational structures, and required competencies while emphasizing the continuing importance of human judgment, trust, explainability, and the interaction between humans and AI. The implication is that AI does not eliminate the need for finance professionals. Instead, it changes the location and nature of professional expertise within the finance process.

This change can be understood as a movement away from the traditional concentration of finance work on information production and toward greater emphasis on information interpretation and decision support. As AI increasingly automates the collection, classification, summarization, and initial analysis of information, finance professionals may devote more time to questioning assumptions, investigating exceptions, assessing uncertainty, and advising management.

Such a transition could strengthen the strategic role of the finance function. The CFO AI Navigator 2026 argues that AI should enable finance to move beyond repetitive operational work and increase its capacity for business partnering and strategic contribution (Parnova Consulting, 2026). However, this outcome is not automatic. Time released through automation does not necessarily become strategic capacity unless organizations deliberately redesign roles, processes, and expectations.

The redesign of work is therefore a central component of human-AI collaboration. Simply providing employees with AI tools while retaining existing job structures may result in inconsistent adoption, duplicated effort, or the emergence of informal and poorly governed uses of AI. A more effective approach requires organizations to determine explicitly which activities AI should perform, which outputs require review, who has authority to override AI recommendations, and where final accountability remains.

This issue becomes particularly significant as AI systems become more autonomous. A generative AI system that prepares a draft report presents a different organizational challenge from an agentic system capable of initiating workflows or coordinating multiple finance activities. As autonomy increases, human involvement should not necessarily disappear. Instead, the form of oversight may change from direct execution to monitoring, review, approval, and intervention.

Eisikovits, Johnson, and Markelevich (2024) identify related concerns regarding professional trust, accountability, bias, data governance, and potential deskilling in accounting and auditing. These concerns demonstrate that human oversight is not merely a technical safeguard. It is also a mechanism for preserving professional responsibility and ensuring that AI-generated outputs are interpreted within an appropriate ethical, organizational, and regulatory context.

The risk of deskilling deserves particular attention. If AI routinely performs tasks that traditionally help employees develop professional expertise, finance organizations may inadvertently weaken their long-term capacity to evaluate complex situations independently. Junior professionals who rely extensively on AI-generated analyses may become more productive in the short term but may have fewer opportunities to develop the underlying analytical capabilities required to challenge AI outputs. Over time, this could create a paradox: the organization becomes increasingly dependent on AI while simultaneously losing some of the human expertise required to supervise it effectively.

CFOs and finance leaders should therefore treat capability development as an essential component of AI implementation. Employees require more than technical training in how to operate AI tools. They need the ability to assess output quality, identify potential errors, understand model limitations, recognize situations requiring escalation, and maintain professional skepticism. AI literacy should therefore complement, rather than replace, core financial and accounting expertise.

Dong, Stratopoulos, and Wang (2024) similarly demonstrate that the growing use of large language models in accounting and finance has implications not only for specific applications but also for professionals and organizations. Stratopoulos and Wang (2025) further argue that AI is changing the direction of accounting research and increasing the importance of combining domain expertise with AI-related capabilities. These findings reinforce the conclusion that the future finance professional will require a broader combination of financial knowledge, analytical judgment, technological understanding, and AI literacy.

Human-AI collaboration should consequently be understood as an organizational capability rather than a temporary implementation arrangement. Effective collaboration requires clearly defined responsibilities, appropriate decision rights, transparent escalation procedures, and sufficient employee capability to evaluate AI outputs. The organization must also recognize that the appropriate balance between human and machine involvement will vary according to the task.

Routine and highly standardized activities may justify substantial automation. Analytical activities involving uncertainty may be more appropriately structured around AI-supported professional judgment. Highly material decisions, complex accounting judgments, or actions with significant regulatory or financial consequences may require explicit human review and approval even when AI provides extensive analytical support.

This differentiated approach is consistent with the broader evidence concerning the task-dependent nature of AI performance. Noy and Zhang (2023) demonstrate that generative AI can produce substantial productivity and quality improvements in specific forms of professional knowledge work. Dell'Acqua et al. (2025), however, demonstrate that such benefits cannot be assumed across all tasks. The managerial challenge is therefore to develop a realistic understanding of the boundary between activities in which AI can operate reliably and those in which human expertise remains particularly important.

For finance functions, the resulting principle is clear: AI should augment professional capacity where it can reliably improve speed, scale, consistency, or analytical breadth, while human professionals retain responsibility for contextual interpretation, material judgment, accountability, and decisions with significant consequences.

This principle also has implications for performance measurement. The benefits of AI should not be assessed solely by the number of tasks automated or the number of employee hours reduced. A more meaningful evaluation considers whether human-AI collaboration improves the performance of the overall finance process. If AI allows employees to focus on higher-value activities while maintaining or improving quality and control, the resulting organizational value may exceed that of a purely automated process.

Ultimately, the most important transformation may therefore be neither technological nor purely economic. It may be the redesign of professional work. AI changes what finance professionals spend their time doing and, potentially, what organizations expect the finance function to contribute. The strategic objective should not be to remove humans from finance processes wherever technology permits it. It should be to design a finance operating model in which human judgment and AI capabilities reinforce one another.

The future of AI-enabled finance is consequently unlikely to be defined by simple substitution. It is more likely to depend on the organization's ability to create effective forms of complementarity between technological capability and professional expertise. For CFOs, the challenge is therefore to determine not how many human tasks can be automated, but where human judgment creates the greatest value and how AI can strengthen rather than undermine that judgment.

8. AI Governance as an Operating Capability

The CFO AI Navigator 2026 identifies data privacy, model validation, explainability, and governance as central requirements for the responsible implementation of AI in finance (Parnova Consulting, 2026). The academic literature strongly reinforces this proposition. As AI systems become increasingly embedded in organizational processes and financial decision-making, governance cannot be treated as a peripheral compliance exercise. It must become an operating capability: a set of organizational structures, processes, skills, controls, and decision rights that enable AI to be deployed, monitored, challenged, and adapted throughout its lifecycle.

This perspective is particularly important because AI creates governance challenges that differ from those associated with conventional information systems. Traditional enterprise software generally operates according to predefined rules and produces predictable outputs from structured inputs. AI systems, by contrast, may generate probabilistic outputs, learn from or be influenced by changing data, produce unexpected results, and operate with varying levels of transparency. Generative AI and agentic systems introduce further complexity because they may interpret unstructured information, generate novel outputs, or coordinate sequences of actions across organizational systems.

The appropriate governance response is therefore not simply to add an approval process before an AI application is deployed. Effective governance must address the full lifecycle of the system, from use-case selection and data preparation through implementation, monitoring, modification, and eventual retirement.

The research of Almeida and Santos Júnior (2025) supports this organizational perspective. Their empirical study of AI governance across 28 public organizations demonstrates that governance involves multiple organizational layers and that capability development is associated with more advanced governance practices. In particular, the study emphasizes that AI governance cannot be reduced to technical controls. Decision-makers, developers, users, and other stakeholders require appropriate knowledge and responsibilities if AI is to be governed effectively.

Hadley, Blatecky, and Comfort (2025) similarly demonstrate the importance of organizational structures for responsible AI governance. Their study of algorithm review boards highlights the importance of leadership support and the integration of governance mechanisms into existing organizational processes. The implication is that governance mechanisms are unlikely to be effective when they operate as isolated committees disconnected from the systems through which organizations make decisions, manage risks, and allocate responsibility.

For finance functions, these findings have direct relevance. Finance already operates within established systems of internal control, risk management, audit, regulatory compliance, financial reporting, and managerial accountability. AI governance should therefore be integrated with these existing mechanisms rather than established as an entirely separate structure. At the same time, AI introduces specific risks that may require new forms of expertise, monitoring, and challenge.

A robust finance AI governance framework should therefore address at least six interconnected dimensions: data governance, model governance, explainability, human accountability, security and confidentiality, and regulatory compliance.

8.1 Data Governance

AI systems are fundamentally dependent on the data used to train, configure, retrieve information for, or otherwise support their operation. In finance, these data may include transactional information, customer and supplier records, financial statements, management reports, forecasts, employee information, tax data, and strategically sensitive corporate information. The reliability of AI outputs is therefore inseparable from the quality and governance of the underlying data (Abbas, 2026; Dong et al., 2024; Stratopoulos & Wang, 2025).

Key issues include data ownership, data lineage, master-data quality, access controls, confidentiality, retention, and provenance. Organizations must understand where relevant information originates, how it has been transformed, who is authorized to access it, and whether it is appropriate for use in a particular AI application. These requirements form part of broader AI governance, which emphasizes organizational accountability, oversight, controls, and the responsible management of AI systems (Almeida & Santos Júnior, 2025; Hadley et al., 2025; Stratopoulos & Wang, 2025).

The principle of data lineage is especially important for finance because AI-generated conclusions may influence material decisions. If an AI system identifies an anomaly, produces a forecast, or generates a management explanation, finance professionals should be able to determine the information on which that output was based. Without an adequate understanding of data sources and transformations, it becomes difficult to challenge the reliability of the result and exercise appropriate professional judgment (Abbas, 2026; Eisikovits et al., 2024; Stratopoulos & Wang, 2025). This is particularly relevant as AI becomes increasingly integrated into accounting and finance processes, where the explainability, reliability, and professional oversight of AI-supported outputs remain important considerations (Dong et al., 2024; Stratopoulos & Wang, 2025).

Poor data quality can therefore turn sophisticated AI into sophisticated error generation. AI can process information at greater speed and scale than human users, but it can also reproduce, amplify, or conceal underlying data problems. An inaccurate master-data record or systematically incomplete data source may therefore affect a large number of AI-supported outputs. Research on AI and knowledge-worker productivity also indicates that the benefits of generative AI are not uniform across tasks and depend on the interaction between AI capabilities, the task, and the quality of human-AI collaboration (Noy & Zhang, 2023; Brynjolfsson et al., 2025; Dell'Acqua et al., 2025).

This challenge is not limited to numerical accuracy. Finance AI may also rely on unstructured information, including contracts, reports, emails, policies, and management commentary. The organization must therefore consider whether these materials are complete, current, appropriately authorized, and relevant to the intended use case. The increasing application of generative AI in accounting and finance makes these issues particularly important because AI systems may draw on and transform diverse forms of organizational information (Dong et al., 2024; Abbas, 2026; Stratopoulos & Wang, 2025).

Data governance should consequently be treated as a prerequisite for reliable AI rather than as a technical issue to be resolved after implementation. Effective AI adoption requires organizations to establish appropriate governance mechanisms, accountability, review processes, and controls around the information on which AI systems depend (Almeida & Santos Júnior, 2025; Hadley et al., 2025). From a finance leadership perspective, this also aligns with emerging practitioner guidance emphasizing the need for CFOs, finance leaders, and boards to address data, governance, risk, and controls when adopting AI (Parnova Consulting, 2026).

8.2 Model Governance

The second dimension concerns the governance of the AI models and systems themselves. Model governance should provide a structured basis for determining whether an AI application is sufficiently reliable for its intended purpose and whether its performance remains acceptable over time. This requires organizations to establish appropriate oversight, accountability, validation, and monitoring mechanisms that reflect the risks associated with the particular AI application (Almeida & Santos Júnior, 2025; Hadley et al., 2025; Stratopoulos & Wang, 2025).

Relevant controls may include independent or appropriately structured validation, ongoing performance monitoring, documentation of assumptions and limitations, version control, testing against known or representative cases, and defined escalation procedures when unexpected performance occurs. The appropriate intensity of these controls should depend on the materiality and risk of the use case. A low-risk application used to generate an initial draft may require a different level of validation from an AI system that influences financial reporting, liquidity management, fraud detection, or significant resource-allocation decisions. This risk-based approach is consistent with the broader literature on responsible AI governance, which emphasizes the need for organizational oversight and mechanisms capable of reviewing AI systems according to their potential risks and impacts (Hadley et al., 2025; Almeida & Santos Júnior, 2025). Within accounting specifically, the risks and opportunities associated with AI also require consideration of the extent to which human professionals should retain responsibility for evaluating and using AI-generated outputs (Eisikovits et al., 2024; Abbas, 2026).

Monitoring is particularly important because AI performance cannot necessarily be assumed to remain constant. Data may change, organizational processes may evolve, user behaviour may alter the way the system is used, and external conditions may reduce the relevance of historical patterns. In such circumstances, a model that performed adequately during implementation may become less reliable over time. Consequently, AI governance should extend beyond initial implementation and incorporate continuing review of system performance, limitations, and appropriateness for its intended use (Abbas, 2026; Hadley et al., 2025; Stratopoulos & Wang, 2025).

The task-dependent nature of AI performance identified by Dell'Acqua et al. (2025) reinforces the importance of ongoing model governance. Their research demonstrates that AI performance can vary substantially depending on whether a task falls within the technology's effective capability frontier. This finding is consistent with wider evidence that the productivity and effectiveness of generative AI depend on the nature of the task and the interaction between human users and AI capabilities (Noy & Zhang, 2023; Brynjolfsson et al., 2025; Dell'Acqua et al., 2025). For finance organizations, this means that governance should not rely solely on average performance measures. Monitoring should also identify the types of cases in which the system performs poorly and establish clear boundaries for its use. Such an approach is particularly important where AI is applied to accounting and financial activities because the suitability of an AI system may differ substantially across tasks and decision contexts (Dong et al., 2024; Stratopoulos & Wang, 2025).

Abbas (2026) similarly identifies questions concerning accuracy, trust, explainability, and the relationship between human and machine judgment as central issues in the continuing development of AI in management accounting. Model governance is therefore not simply concerned with technical performance. It must also establish whether the system is appropriate for the organizational decisions in which it is being used and whether human professionals retain sufficient oversight to evaluate its outputs critically (Abbas, 2026; Eisikovits et al., 2024; Stratopoulos & Wang, 2025).

8.3 Explainability and Contestability

Explainability is a third critical dimension of AI governance. In finance, users may need to understand why an AI system produced a particular recommendation, forecast, classification, or conclusion and what evidence supports that output. This requirement becomes increasingly important as the materiality and potential consequences of the decision increase. In accounting and finance, concerns about the reliability, trustworthiness, explainability, and appropriate use of AI-generated outputs are therefore closely connected to the exercise of professional judgment (Abbas, 2026; Eisikovits et al., 2024; Stratopoulos & Wang, 2025).

Explainability does not necessarily require complete disclosure of every technical feature or internal model mechanism. In many cases, particularly where complex commercial AI models are used, complete technical transparency may not be feasible. The relevant governance objective is instead sufficient explainability for meaningful human challenge and accountability. This reflects the broader governance perspective that effective AI oversight requires mechanisms through which organizations can understand, evaluate, and hold systems and their users accountable for AI-supported outcomes (Almeida & Santos Júnior, 2025; Hadley et al., 2025).

A finance professional should be able, to an appropriate degree, to determine the basis of an AI-generated output, understand the major assumptions or limitations involved, identify relevant sources of uncertainty, and challenge conclusions that appear inconsistent with business knowledge or other available evidence. This is particularly important for applications involving forecasts, accounting judgments, risk assessments, anomaly detection, or other outputs that may influence material decisions. The accounting literature emphasizes that AI should complement rather than displace professional judgment, making the ability of finance professionals to critically evaluate AI outputs an important component of responsible AI adoption (Abbas, 2026; Eisikovits et al., 2024). Research on AI in accounting and finance likewise highlights the need to consider how AI-generated information is interpreted, evaluated, and incorporated into professional decision-making (Dong et al., 2024; Stratopoulos & Wang, 2025).

Explainability should therefore be understood together with contestability. An AI output is governed effectively only when authorized individuals can challenge it and when there is an established process for responding to that challenge. The organization should not create a situation in which AI recommendations become de facto decisions because employees lack the information, expertise, or authority required to question them. This is consistent with responsible AI governance approaches that emphasize organizational accountability, human oversight, and formal mechanisms for reviewing or challenging AI-supported decisions (Hadley et al., 2025; Almeida & Santos Júnior, 2025). In accounting and auditing, such safeguards are particularly relevant because excessive reliance on AI could otherwise weaken rather than support professional skepticism and judgment (Eisikovits et al., 2024).

This issue is closely related to the concerns identified by Eisikovits, Johnson, and Markelevich (2024), who emphasize the risks and opportunities associated with AI adoption in accounting and auditing, including issues of professional trust, bias, accountability, and professional judgment. If professionals cannot reasonably assess the basis of an AI-generated recommendation, their ability to exercise meaningful judgment may be weakened rather than strengthened (Eisikovits et al., 2024; Abbas, 2026). More broadly, this supports the view that explainability should not be treated merely as a technical characteristic of an AI model, but as a governance capability that enables appropriate human oversight and accountability (Hadley et al., 2025; Stratopoulos & Wang, 2025).

8.4 Human Accountability

Perhaps the most fundamental governance requirement is the clear allocation of human accountability. An organization must establish who is responsible when an AI-supported decision is incorrect, harmful, or inconsistent with applicable policy or regulation. Effective AI governance therefore requires clearly defined responsibilities and organizational oversight rather than treating responsibility as an inherent property of the technology itself (Almeida & Santos Júnior, 2025; Hadley et al., 2025).

The statement “AI decided” is not an acceptable governance structure for a finance function. AI systems do not eliminate organizational responsibility. They may influence how decisions are made, but accountability for financial outcomes remains with identifiable individuals and organizational functions. This is particularly important in accounting and auditing, where the introduction of AI raises questions concerning professional judgment, accountability, trust, and the appropriate relationship between human and machine decision-making (Eisikovits et al., 2024; Abbas, 2026).

The allocation of accountability should therefore be defined before deployment. Depending on the application, the organization may need to distinguish among the business owner of the process, the technical owner of the AI system, the data owner, the individual responsible for validating outputs, and the person with authority to approve consequential actions. These responsibilities should not be assumed to overlap automatically. Establishing explicit roles and oversight mechanisms is consistent with research on organizational AI governance, which emphasizes the importance of assigning responsibility across organizational levels and creating formal structures through which AI systems can be monitored and reviewed (Almeida & Santos Júnior, 2025; Hadley et al., 2025).

The appropriate structure will depend on the nature of the AI application. A system that assists with routine document classification may permit extensive automation with limited human intervention. A system that influences material accounting judgments or significant financial decisions should involve more explicit human review and accountability. This risk- and context-dependent approach is consistent with the accounting literature, which highlights that the opportunities and risks of AI adoption vary according to the nature of the task and the extent to which professional judgment remains necessary (Eisikovits et al., 2024; Abbas, 2026; Stratopoulos & Wang, 2025).

This principle is particularly important as AI systems become more autonomous. Generative AI can produce recommendations without executing actions, whereas agentic systems may potentially coordinate workflows and initiate actions across multiple systems. As autonomy increases, governance must establish not only who reviews outputs but also which actions the system is authorized to perform independently, which require approval, and which are prohibited altogether. The need for structured oversight becomes especially significant as AI systems are integrated more deeply into organizational processes, increasing the importance of defined controls, accountability, and human supervision (Hadley et al., 2025; Almeida & Santos Júnior, 2025; Parnova Consulting, 2026).

The research on organizational AI governance supports this approach. Almeida and Santos Júnior (2025) demonstrate that governance depends on the allocation of responsibilities across organizational levels, while Hadley et al. (2025) emphasize the importance of structured organizational oversight. For finance, this implies that accountability should be embedded within the normal architecture of management and control rather than delegated ambiguously to a technology provider or AI development team. This is also consistent with the broader accounting literature, which emphasizes the continuing importance of human responsibility and professional judgment when AI is incorporated into accounting and auditing activities (Eisikovits et al., 2024; Abbas, 2026).

8.5 Security and Confidentiality

Security and confidentiality represent a fifth critical dimension. Finance functions process some of the most sensitive information within an organization, including employee data, customer and supplier information, transactions, financial forecasts, tax positions, pricing information, strategic plans, and potentially market-sensitive information. The sensitivity of these information assets makes security, confidentiality, and appropriate controls important considerations when AI is incorporated into accounting and finance processes (Eisikovits et al., 2024; Abbas, 2026; Stratopoulos & Wang, 2025).

AI implementation must therefore incorporate information-security and confidentiality requirements from the beginning. The organization must understand what information is provided to an AI system, where that information is processed, whether it is retained, who can access it, and whether its use creates unintended exposure. These requirements form part of broader AI governance because responsible implementation depends not only on the technical capabilities of an AI system but also on organizational controls governing its use and the information to which it has access (Almeida & Santos Júnior, 2025; Hadley et al., 2025).

Generative AI creates particular challenges because users may interact with systems through natural-language prompts. Employees may inadvertently provide sensitive information to an AI application without fully understanding how that information is processed or retained. Governance should therefore address not only system-level security but also employee behaviour, training, and acceptable-use policies. The increasing use of generative AI in accounting and finance makes such controls particularly relevant because AI applications may be integrated directly into everyday professional workflows and may process a broad range of organizational information (Dong et al., 2024; Abbas, 2026).

This reinforces the broader findings of the AI governance literature that responsible AI depends on organizational capability. Technical security controls are essential, but employees must also understand what information can and cannot be used with particular AI systems. Training and clearly defined procedures are therefore essential components of governance rather than supplementary activities. Effective AI governance requires organizational structures, responsibilities, oversight, and processes that enable employees to use AI systems consistently with organizational requirements (Almeida & Santos Júnior, 2025; Hadley et al., 2025).

The significance of these issues is also consistent with the concerns identified by Eisikovits et al. (2024), particularly regarding data ownership, professional trust, and the risks associated with the integration of AI into accounting and auditing activities. In finance, confidentiality failures can have direct regulatory, financial, contractual, and reputational consequences. These risks reinforce the need for finance organizations to integrate AI governance with existing information-security, risk-management, and internal-control arrangements rather than treating AI security as a separate technological concern (Eisikovits et al., 2024; Almeida & Santos Júnior, 2025; Parnova Consulting, 2026).

8.6 Regulatory and Cross-Border Compliance

The sixth dimension concerns regulatory and legal compliance. AI governance for finance cannot be designed independently of the legal and regulatory environment in which the organization operates. Relevant requirements may arise from financial regulation, data-protection law, corporate governance requirements, contractual obligations, sector-specific regulation, and internal organizational policies. This is consistent with the broader AI governance literature, which emphasizes that responsible AI requires organizations to integrate legal, regulatory, accountability, and oversight considerations into the design and operation of AI systems (Almeida & Santos Júnior, 2025; Hadley et al., 2025).

For Swiss companies, governance must therefore account for applicable Swiss requirements while also considering cross-border obligations. Switzerland currently does not have a single overarching statute specifically regulating AI; instead, AI is addressed through the existing legal framework alongside ongoing work toward new AI-specific regulation. The Federal Council has instructed the Federal Department of Justice to prepare a consultation draft by the end of 2026, with particular attention to transparency, data protection, non-discrimination, and supervision. Swiss data-protection requirements already apply directly to AI-supported data processing under the Federal Act on Data Protection (FADP). The Federal Data Protection and Information Commissioner also emphasizes transparency concerning the purpose, functionality, and data sources of AI-based processing and requires data-protection impact assessments in high-risk cases.

Organizations operating internationally or serving customers and counterparties in multiple jurisdictions may face overlapping legal and regulatory expectations. European regulatory developments may also be relevant to Swiss organizations depending on their activities, operations, and relationships within European markets. The Swiss authorities themselves recognize the importance of considering international developments, including the Council of Europe's AI Convention and the EU AI Act, when developing Switzerland's regulatory approach.

The central governance principle is that regulatory compliance should not be assessed only after an AI system has been selected or implemented. Legal, regulatory, and contractual considerations can affect the feasibility of a proposed use case and should therefore be incorporated into the initial assessment and business case. This lifecycle approach is consistent with responsible AI governance research, which emphasizes the need to establish governance mechanisms and organizational responsibilities as part of AI implementation rather than treating governance as a separate downstream activity (Almeida & Santos Júnior, 2025; Hadley et al., 2025).

This lifecycle perspective is also consistent with the CFO AI Navigator 2026, which emphasizes the importance of designing governance early rather than treating it as a final-stage compliance activity (Parnova Consulting, 2026). An AI initiative may be technically attractive but economically or operationally unsuitable if the requirements for data protection, model validation, auditability, or human oversight fundamentally alter the cost or complexity of implementation. This reinforces the broader finding that AI adoption involves both potential benefits and organizational risks that must be evaluated in the context of the particular application (Eisikovits et al., 2024; Abbas, 2026).

Governance as a Dynamic Organizational Capability

Taken together, these six dimensions demonstrate why AI governance should be understood as an operating capability rather than a static set of policies. Effective governance requires an organization to make repeated decisions throughout the AI lifecycle: whether a use case should proceed, whether available data are appropriate, whether the system is sufficiently reliable, whether outputs require human review, whether performance has changed, and whether the system should be modified, restricted, or withdrawn. This dynamic conception is consistent with research emphasizing continuing oversight, organizational accountability, and the need to adapt governance arrangements to the characteristics and risks of AI applications (Almeida & Santos Júnior, 2025; Hadley et al., 2025; Stratopoulos & Wang, 2025).

This dynamic perspective is particularly important as AI technologies evolve. Governance arrangements that are sufficient for conventional automation may not be adequate for generative AI, and frameworks developed for generative AI may require further adaptation as more autonomous or agentic systems are deployed. The increasing diversity of AI applications reinforces the need for governance frameworks that can be adapted to different technologies, tasks, and organizational contexts (Dong et al., 2024; Stratopoulos & Wang, 2025). The task-dependent nature of AI performance identified by Dell'Acqua et al. (2025) further suggests that governance cannot rely on a single assessment of AI capability but should consider how performance varies across different tasks and circumstances.

The finance function should play a significant role in this process because many of the most important governance questions concern value, materiality, risk, accountability, and internal control. However, governance should remain multidisciplinary. CFOs and finance leaders require collaboration with IT, cybersecurity, legal, compliance, data specialists, risk management, internal audit, and relevant business stakeholders. This multidisciplinary approach reflects the organizational nature of AI governance identified by Almeida and Santos Júnior (2025) and Hadley et al. (2025), while the accounting literature similarly emphasizes the continuing importance of professional judgment and organizational oversight when AI is incorporated into finance and accounting activities (Eisikovits et al., 2024; Abbas, 2026).

The resulting governance model should therefore be business-led, multidisciplinary, and proportionate to risk. Low-risk AI applications may require relatively simple controls, while applications that influence material financial decisions or operate with greater autonomy should be subject to more rigorous validation, monitoring, and human oversight. A proportionate approach is consistent with the risk-sensitive orientation of responsible AI governance and with the accounting literature's emphasis on considering the specific risks and opportunities associated with different AI applications (Hadley et al., 2025; Almeida & Santos Júnior, 2025; Eisikovits et al., 2024).

Ultimately, the purpose of governance is not to prevent AI adoption. Excessive or poorly designed controls can themselves become an obstacle to innovation and organizational learning. The objective is instead to create conditions in which AI can be deployed with sufficient confidence that the organization can capture its benefits while managing its risks. This is consistent with the broader literature, which identifies AI as presenting both significant opportunities and risks and therefore requiring governance mechanisms capable of supporting responsible adoption rather than simply restricting technological use (Eisikovits et al., 2024; Abbas, 2026).

This leads to a central conclusion: effective AI governance is not a constraint placed on transformation from outside the business. It is one of the capabilities that makes sustainable AI transformation possible. Without reliable data, appropriate validation, meaningful explainability, clear accountability, effective security, and regulatory awareness, AI may create short-term experimentation but limited sustainable value. The governance literature supports this broader organizational perspective by treating accountability, oversight, risk management, and responsible implementation as integral components of AI adoption (Almeida & Santos Júnior, 2025; Hadley et al., 2025).

For CFOs, governance should therefore be treated as part of the operating model of AI-enabled finance. The relevant objective is not merely to ensure that individual AI systems comply with policy. It is to build an organizational capability that allows finance to identify valuable opportunities, deploy AI responsibly, monitor performance continuously, challenge inappropriate outputs, and scale successful applications without compromising control, accountability, or professional judgment (Abbas, 2026; Eisikovits et al., 2024; Stratopoulos & Wang, 2025; Parnova Consulting, 2026).

9. Organizational Change and the Transformation of the Finance Profession

Technology implementation without corresponding organizational redesign is unlikely to deliver the full value of artificial intelligence. The introduction of AI into finance does not simply alter the tools available to employees; it changes how information is produced, how decisions are supported, how professional responsibilities are distributed, and which capabilities become strategically important. AI adoption should therefore be understood as a workforce and organizational transformation rather than merely as a technology implementation.

This perspective is strongly supported by the emerging accounting literature. Abbas (2026), in a comprehensive review of management accounting and AI research, identifies significant implications for automation, upskilling, reskilling, deskilling, professional roles, organizational structures, and multidisciplinary collaboration. The review suggests that AI is not merely automating isolated accounting activities. It is contributing to broader changes in the boundaries and composition of management accounting work.

Dong, Stratopoulos, and Wang (2024) similarly demonstrate that the growing development and adoption of large language models has implications extending beyond individual technical applications. Their review of ChatGPT research in accounting and finance identifies applications across a wide range of accounting and financial activities while also drawing attention to the implications of AI for professionals, organizations, and research. Stratopoulos and Wang (2025) further argue that the increasing importance of AI is reshaping the research agenda of accounting and increasing the importance of combining domain knowledge with technological understanding.

Taken together, this literature suggests that AI adoption should be accompanied by a deliberate finance workforce strategy. Organizations should not assume that the skills required for an AI-enabled finance function will emerge automatically once employees are given access to new technologies. Instead, CFOs and finance leaders must consider how roles, responsibilities, training, career development, performance expectations, and organizational structures should change.

Traditionally, finance roles have emphasized capabilities such as accounting knowledge, financial reporting, spreadsheet modelling, reconciliation, internal controls, variance analysis, and the preparation of management information. These capabilities remain important. However, as AI increasingly automates elements of information collection, classification, summarization, and analysis, the relative importance of other capabilities is likely to increase.

The AI-enabled finance professional requires stronger data literacy, including an understanding of data quality, provenance, limitations, and appropriate interpretation. Employees also require AI literacy: the ability to understand the basic capabilities and limitations of AI systems, recognize where AI is appropriate, formulate effective requests and workflows, and identify circumstances in which AI outputs should not be trusted without further investigation.

Model interpretation and critical evaluation are also becoming increasingly important. Finance professionals do not necessarily need to become AI engineers, but they increasingly need the ability to assess whether an AI-generated output is reasonable, understand the assumptions and evidence on which it is based, recognize potential sources of error, and challenge conclusions that are inconsistent with financial or business knowledge.

The transformation also increases the importance of process design. When AI automates or augments individual tasks, organizations must reconsider the structure of the broader process. The relevant question is no longer simply how employees can perform existing activities faster. It is whether the process itself should be redesigned. Financial close, forecasting, reporting, reconciliations, and management analysis may all require new workflows in which humans and AI systems perform different activities and interact at different points.

These changes also increase the importance of automation and AI governance as professional competencies. Finance employees may increasingly be required to determine which activities can be delegated to AI, which require human review, what types of errors are acceptable, when exceptions should be escalated, and how AI-supported decisions should be documented. These capabilities connect directly with the governance challenges identified by Almeida and Santos Júnior (2025) and Hadley, Blatecky, and Comfort (2025), both of which emphasize that effective AI governance depends on organizational structures, leadership, and the development of relevant capabilities across multiple groups.

At the same time, the potential automation of routine activities may strengthen the importance of business partnering, communication, and storytelling. If AI reduces the time required to prepare information, finance professionals may have greater capacity to focus on explaining what that information means, evaluating strategic alternatives, communicating uncertainty, and supporting management decisions. This development is consistent with the strategic role envisioned in the CFO AI Navigator 2026, which presents AI as an opportunity for finance to move beyond repetitive operational activities and contribute more actively to organizational steering and decision-making (Parnova Consulting, 2026).

The transformation is therefore less about the elimination of accountants and finance professionals than about a change in the composition of finance work. Routine information processing may decline in relative importance, while activities requiring judgment, contextual understanding, challenge, communication, governance, and strategic interpretation may become more significant.

This does not mean, however, that all AI-related change will be positive or that professional expertise will automatically become more valuable. Abbas (2026) specifically identifies the potential for deskilling alongside upskilling and reskilling. This distinction is important because AI can simultaneously improve short-term productivity and weaken the development of long-term professional expertise.

For example, junior finance professionals have traditionally developed judgment through repeated exposure to reconciliations, financial statements, forecasting exercises, and detailed analytical work. If AI performs an increasing proportion of these activities, employees may have fewer opportunities to develop the foundational knowledge required to recognize when an AI output is incorrect. The organization could therefore face a paradox in which greater reliance on AI reduces the availability of the human expertise necessary to supervise that technology effectively.

The findings of Brynjolfsson, Li, and Raymond (2025) are particularly relevant in this context. Their research demonstrates that AI can produce disproportionately large productivity gains for less-experienced workers. This suggests that AI may help newer employees perform tasks that would otherwise require greater experience. Such capability can improve productivity and potentially accelerate learning. However, it also raises an important organizational question: does AI-assisted performance represent genuine capability development, or does it create dependence on a technological system whose outputs employees are unable to evaluate independently?

The answer is likely to depend on how organizations design learning and work. AI should not simply be used as a substitute for professional development. Finance organizations should consider whether AI can instead support learning by making expert reasoning more accessible while continuing to require employees to understand, challenge, and independently assess important conclusions.

This issue becomes particularly important because generative AI can produce highly plausible but incorrect outputs. Unlike conventional calculation errors, which may be relatively visible, generative AI errors can be presented in fluent and persuasive language. A well-written financial explanation may therefore appear credible even when it contains incorrect assumptions, fabricated information, or inappropriate interpretations.

Generative AI consequently creates an important professional paradox: the easier it becomes to produce financial analysis, the more important expert review may become. When the cost of producing a first draft approaches zero, the scarce capability may shift from generating information to evaluating whether that information is accurate, relevant, and appropriate for decision-making.

This is consistent with the concerns raised by Eisikovits, Johnson, and Markelevich (2024), who identify risks associated with AI adoption in accounting and auditing, including professional trust, bias, governance, data ownership, and the possible erosion of professional capabilities. The central challenge is not simply to teach finance professionals how to use AI. It is to preserve and strengthen the professional skepticism required to challenge AI-generated outputs.

Noy and Zhang (2023) provide additional insight into this issue. Their experimental findings demonstrate that generative AI can improve productivity and output quality in professional knowledge work. However, productivity improvements do not eliminate the need for human judgment regarding the appropriate use of the resulting outputs. The organizational value of AI therefore depends not only on whether employees can generate work more quickly but also on whether they understand how and when to rely on AI assistance.

Dell'Acqua et al. (2025) reinforce this conclusion through evidence that AI effectiveness is task-dependent. Their research demonstrates that performance may improve when individuals use AI within the technology's effective capability frontier but may deteriorate when users rely on it outside that frontier. For the finance profession, this means that one of the most important emerging capabilities is the ability to recognize the boundaries of AI competence.

Employees must increasingly know when an AI-generated forecast should be accepted as useful input, when it should be challenged, and when the underlying problem falls outside the system's reliable capabilities. This requires both technical awareness and deep domain expertise. The finance professional of the future may therefore be distinguished less by the ability to produce every analysis manually and more by the ability to determine which analyses can safely be supported by AI and how the resulting outputs should be interpreted.

These developments also have implications for organizational structure. Abbas (2026) emphasizes that AI can change professional boundaries and encourage multidisciplinary collaboration. AI-enabled finance therefore requires closer interaction among finance, information technology, data management, cybersecurity, risk, legal, compliance, and internal audit functions.

The resulting organizational model may be less functionally isolated than traditional finance structures. Finance professionals will require sufficient technological understanding to participate meaningfully in AI decisions, while technology specialists will require sufficient understanding of finance processes, controls, and materiality. Effective AI transformation consequently depends on creating a shared language across disciplines.

This does not imply that every finance employee must become a data scientist or AI specialist. Instead, finance organizations should develop differentiated capability profiles. Some employees may require advanced technical expertise in data, analytics, or AI systems. Others may primarily require sufficient AI literacy to use tools responsibly and evaluate outputs critically. Senior finance leaders may require the capability to make strategic decisions about AI investment, governance, risk, and operating-model design.

Training should therefore be aligned with roles and responsibilities. A standardized AI training programme may provide useful basic knowledge, but it is unlikely to address the full range of organizational needs. Employees responsible for material financial decisions require different capabilities from those using AI primarily for document summarization or routine reporting. Similarly, employees involved in AI governance require additional knowledge concerning risk assessment, accountability, validation, and escalation.

The CFO AI Navigator 2026 therefore correctly identifies organizational change and skills development as integral components of implementation rather than activities that occur after a technical deployment is complete (Parnova Consulting, 2026). This principle is supported by the broader evidence on AI adoption: technological capability produces value only when organizations adapt their processes, roles, and capabilities to use that technology effectively.

The transformation of the finance profession can thus be understood as the integration of several complementary capabilities. Domain expertise remains fundamental because AI outputs require financial and business interpretation. Data literacy becomes increasingly important because the quality and usefulness of AI applications depend heavily on the quality, availability, and context of underlying information. AI fluency is required to understand the technology’s capabilities and limitations, while professional judgment remains essential for dealing with uncertainty, ambiguity, materiality, and accountability (Abbas, 2026; Eisikovits, Johnson and Markelevich, 2024). Finally, business-partnering capabilities become increasingly valuable as finance professionals devote greater attention to interpreting information and supporting organizational decisions (Stratopoulos and Wang, 2025). The literature therefore suggests that AI adoption is likely to broaden rather than eliminate the capability requirements of the finance profession.

The interaction among these capabilities is more important than any single technical skill. An employee with strong AI knowledge but limited understanding of finance may be unable to evaluate the material significance of an output. Conversely, a highly experienced accountant without sufficient AI literacy may struggle to recognize how the technology can be used effectively or where its limitations create risk. The emerging finance profession therefore requires the integration of technological understanding with established professional knowledge and judgment (Abbas, 2026; Dong, Stratopoulos and Wang, 2024). This is particularly important because AI performance varies across tasks and circumstances: understanding where AI is reliable requires users to possess sufficient domain knowledge to recognize inappropriate or low-quality outputs (Dell’Acqua et al., 2026).

The distribution of these capabilities across the workforce may also change as AI becomes more widely adopted. Brynjolfsson, Li and Raymond (2025) find that generative AI can produce particularly strong productivity benefits for less-experienced workers, suggesting that AI may accelerate the development of certain forms of routine expertise. In finance, this could reduce the time required for junior professionals to perform standardized analytical and reporting activities. At the same time, experienced professionals may increasingly concentrate on reviewing AI-generated outputs, resolving exceptions, interpreting complex situations, and exercising professional judgment. AI adoption may therefore change not only the quantity of work performed but also how expertise is developed and distributed within finance teams (Brynjolfsson, Li and Raymond, 2025; Abbas, 2026).

Ultimately, organizational change should be treated as a central mechanism through which AI creates or fails to create value. The implementation of an AI system may change the technical execution of work, but sustainable transformation requires changes in roles, workflows, skills, incentives, governance, and professional expectations. Research on AI governance emphasizes that organizations need appropriate structures for accountability, oversight, and responsible implementation rather than treating AI as an isolated technological intervention (Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025). Similarly, research in accounting indicates that the organizational and professional implications of AI are integral to understanding its long-term impact on the finance function (Dong, Stratopoulos and Wang, 2024; Stratopoulos and Wang, 2025).

For CFOs, the strategic implication is clear: AI transformation requires investment in people as well as technology. Finance leaders should assess not only which processes can be automated or augmented, but also which professional capabilities the future finance function will require and how those capabilities will be developed. This includes investment in AI literacy, data literacy, professional training, change management, and the ability to evaluate and challenge AI-generated outputs (Abbas, 2026; Eisikovits, Johnson and Markelevich, 2024). Governance capabilities are also necessary because employees must understand not only how to use AI but also when its outputs require escalation, validation, or human intervention (Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025).

The most successful AI-enabled finance organizations are therefore unlikely to be those that simply reduce the number of employees performing routine tasks. They are more likely to be those that deliberately redesign finance work so that technological capabilities and professional expertise complement one another. AI may increasingly perform elements of information processing, pattern recognition, routine analysis, and content generation, while finance professionals focus more strongly on interpretation, challenge, governance, communication, and strategic decision support (Abbas, 2026; Stratopoulos and Wang, 2025). Evidence from the productivity literature supports this complementary perspective: AI can enhance human performance when applied to tasks within its effective capability range, but performance may deteriorate when users or organizations extend AI beyond those boundaries (Noy and Zhang, 2023; Dell’Acqua et al., 2026).

This transformation also has implications for how finance organizations define performance and career progression. Traditional finance career structures have often relied on employees developing expertise through repeated exposure to routine analytical, reporting, and control activities. As AI assumes more of these activities, organizations may need to redesign development pathways so that professionals acquire judgment, business understanding, communication, data literacy, and AI-related capabilities earlier in their careers. The productivity gains observed among less-experienced workers suggest that AI can potentially accelerate learning and task performance, but this benefit should not be interpreted as eliminating the need for experienced professionals. Rather, experienced professionals may become increasingly important as reviewers, decision-makers, mentors, and owners of professional standards (Brynjolfsson, Li and Raymond, 2025; Abbas, 2026).

In this sense, the transformation of the finance profession is not a secondary consequence of AI adoption. It is one of the central conditions for realizing AI’s long-term value. The future competitiveness of the finance function will depend not only on the quality of its AI systems, but also on the ability of its professionals to understand, challenge, govern, and strategically apply those systems. This conclusion follows from the broader literature, which presents AI adoption as simultaneously a technological, organizational, and professional transformation (Dong, Stratopoulos and Wang, 2024; Abbas, 2026; Stratopoulos and Wang, 2025). For CFOs, the strategic priority is therefore to develop a finance workforce in which technological fluency reinforces—rather than substitutes for—financial expertise, professional judgment, and business leadership.

10. A Revised Five-Stage CFO AI Transformation Model

Building on the implementation logic of the CFO AI Navigator 2026 and the findings of the emerging academic literature, this paper proposes a revised five-stage model for CFO-led AI transformation: diagnose, prioritize, experiment, redesign, and institutionalize. The model extends the Parnova framework by explicitly incorporating the literature's emphasis on task-level heterogeneity, human-AI complementarity, organizational capabilities, governance, and evidence-based measurement.

The central premise is that AI transformation should be managed as a progression from organizational diagnosis to sustained capability development, rather than as a sequence of technology deployments. Each stage addresses a different source of implementation risk, while the transition between stages depends on evidence that the organization is ready to proceed.

10.1 Stage 1: Diagnose — Establish the AI Readiness Baseline

The first stage is to establish a clear understanding of the existing finance environment. Before selecting technologies or launching pilots, the organization should map its major finance processes, transaction volumes, manual effort, data quality, existing automation, systems architecture, control requirements, and workforce capabilities.

This diagnostic stage should produce an AI readiness map, rather than a technology wish list. The objective is to identify where AI could plausibly create value and where organizational conditions are currently insufficient for successful adoption.

Process assessment is particularly important. The organization should determine which activities are standardized and repetitive, which depend heavily on professional judgment, where bottlenecks occur, and where employees spend significant amounts of time on low-value information processing. This allows AI opportunities to be considered in relation to actual business problems rather than generalized technological capabilities.

Data readiness should be assessed simultaneously. A potentially valuable AI use case may be unsuitable if the necessary information is incomplete, inconsistent, poorly governed, inaccessible, or distributed across incompatible systems. As emphasized by the Parnova framework, AI implementation depends on foundations that include process maturity, data quality, and governance (Parnova Consulting, 2026).

The diagnostic should also examine existing automation. Finance departments may already use rules-based automation, robotic process automation, enterprise software functionality, analytics platforms, or other digital tools. AI should therefore not automatically be considered the first solution to a process problem. In some circumstances, conventional automation or process standardization may provide a more appropriate and lower-risk solution.

Workforce capability is another essential component of the diagnosis. The organization should assess existing levels of accounting expertise, data literacy, AI literacy, process-design capability, and change capacity. Abbas (2026) demonstrates that AI adoption can affect professional roles, skills, organizational structures, and collaboration. Consequently, workforce readiness should be considered alongside technical readiness from the beginning.

The output of Stage 1 should therefore be a structured view of where the organization is ready to use AI, where organizational foundations must first be strengthened, and where AI may not be appropriate at all.

10.2 Stage 2: Prioritize — Select Use Cases According to Value and Readiness

The second stage is to prioritize potential AI applications. Rather than pursuing the largest possible number of opportunities, organizations should identify a limited portfolio of use cases that combine attractive potential value with sufficient feasibility and manageable risk. This portfolio-based approach is consistent with the emerging accounting literature, which emphasizes the need to align AI initiatives with organizational objectives, implementation conditions, and the characteristics of specific accounting and finance tasks (Dong, Stratopoulos and Wang, 2024; Stratopoulos and Wang, 2025).

Candidate applications should be assessed according to their economic value, data readiness, process maturity, implementation complexity, control risk, and strategic relevance. These criteria should be considered together because high potential value alone does not make a use case attractive. Research on AI adoption and governance indicates that organizational readiness, data conditions, implementation capabilities, and governance arrangements can significantly influence whether technically capable AI systems generate sustainable organizational value (Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025). A use case should therefore be evaluated as an organizational intervention rather than simply as a technological opportunity.

Economic value should include both efficiency and decision-related benefits. For transactional processes, relevant benefits may include lower processing effort, reduced error rates, shorter cycle times, and improved throughput. For analytical applications, value may instead arise through improved forecasts, faster scenario analysis, stronger working-capital management, better risk detection, or improved management decisions. This distinction is important because the value of AI extends beyond direct labour savings. Experimental and field evidence demonstrates that generative AI can increase productivity, but the nature and magnitude of these gains depend on the tasks and workers involved (Noy and Zhang, 2023; Brynjolfsson, Li and Raymond, 2025). In accounting, AI can similarly generate value through automation, information processing, analytical support, and decision assistance (Abbas, 2026; Dong, Stratopoulos and Wang, 2024).

Data readiness and process maturity should receive particular attention. A highly attractive use case may fail if the underlying process is poorly standardized or the required data are unreliable. AI performance is shaped by the environment in which the technology operates, including the quality of information, task structure, organizational processes, and user capabilities, rather than being determined solely by the capability of the model (Almeida and Santos Júnior, 2025; Abbas, 2026). This suggests that CFOs should assess whether the organization has sufficiently reliable data, clearly defined processes, and appropriate system infrastructure before committing significant resources to an AI application. Where these foundations are weak, process standardization and data-quality improvements may need to precede AI deployment.

The evidence of heterogeneous AI effects reinforces this principle. Noy and Zhang (2023) demonstrate substantial productivity improvements in particular professional tasks, while Brynjolfsson, Li and Raymond (2025) find considerable variation in the magnitude of productivity gains across workers. Dell’Acqua et al. (2026) go further by demonstrating that AI can improve performance within its effective capability frontier while potentially reducing performance outside it. This “jagged technological frontier” implies that AI should not be assumed to provide uniform benefits across all activities. CFOs should therefore prioritize tasks and processes in which AI has a credible performance advantage, rather than assuming that AI is equally valuable across the finance function (Dell’Acqua et al., 2026).

This task-specific perspective is particularly relevant when comparing automation and analytical use cases. Routine, structured activities may provide relatively predictable opportunities for efficiency improvements, whereas analytical and judgment-intensive activities may offer greater potential decision value but also involve greater uncertainty and dependence on human oversight (Abbas, 2026; Eisikovits, Johnson and Markelevich, 2024). The selection process should therefore distinguish between applications where AI can reliably execute defined activities and those where AI should primarily augment finance professionals by generating information, identifying patterns, or proposing alternatives.

Control risk must also be considered at this stage. A use case that influences financial reporting, accounting judgments, liquidity decisions, or other material outcomes should generally receive greater scrutiny than an application used to prepare a low-risk internal draft. The appropriate governance requirements should therefore be proportional to the consequences of failure. Accounting research identifies accuracy, explainability, professional responsibility, and the appropriate balance between human and machine judgment as important considerations when AI is incorporated into accounting and auditing activities (Eisikovits, Johnson and Markelevich, 2024; Abbas, 2026). More broadly, AI governance research emphasizes the importance of accountability, oversight, review mechanisms, and clearly defined organizational responsibilities (Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025).

Implementation complexity should similarly influence prioritization. Applications requiring extensive systems integration, major process redesign, significant employee retraining, or substantial changes to control structures may have attractive long-term potential but may be unsuitable as initial AI investments. Early initiatives can instead be selected where the organization can establish clear performance baselines, monitor outcomes, and learn from implementation without exposing critical finance processes to disproportionate risk. This staged approach allows organizations to develop AI-related capabilities while limiting the consequences of early experimentation (Almeida and Santos Júnior, 2025; Stratopoulos and Wang, 2025).

The outcome of Stage 2 should therefore be a prioritized AI portfolio with a limited number of high-value opportunities suitable for experimentation. Importantly, the portfolio should include explicit reasons for pursuing particular use cases and for postponing or rejecting others. This creates a disciplined basis for subsequent experimentation and prevents AI adoption from becoming a collection of disconnected technology projects. The prioritization process should also remain dynamic: as data quality improves, organizational capabilities develop, and evidence accumulates regarding AI performance, previously unattractive use cases may become viable while existing applications may need to be redesigned or discontinued (Stratopoulos and Wang, 2025; Abbas, 2026).

Ultimately, Stage 2 should establish a clear principle for AI investment in finance: the organization should prioritize the problems where AI can create demonstrable value under realistic operating and governance conditions, rather than prioritizing AI simply because the technology is available. This shifts the focus from technology-led experimentation toward evidence-based portfolio management and provides the foundation for testing whether individual use cases deliver measurable improvements in finance performance (Abbas, 2026; Stratopoulos and Wang, 2025).

10.3 Stage 3: Experiment — Test Net Organizational Value

The third stage is experimentation. Selected use cases should be tested through carefully designed pilots with explicit baseline metrics, hypotheses, quality thresholds, security requirements, governance ownership, and stop-or-go criteria.

The purpose of a pilot should not simply be to demonstrate that an AI system can technically perform a task. Technical feasibility is only the first question. The more important question is whether AI creates net organizational value within the specific finance process.

This distinction follows directly from the empirical literature. Noy and Zhang (2023) demonstrate that generative AI can improve both productivity and output quality under particular experimental conditions. Brynjolfsson et al. (2025) show that benefits vary among users, while Dell'Acqua et al. (2025) demonstrate that benefits depend on task characteristics. These findings imply that AI pilots should test the technology in the organization's actual operating environment rather than relying on generic claims about AI performance.

Baseline measurement is therefore essential. The organization should establish the performance of the existing process before introducing AI and then assess changes in relevant dimensions such as processing time, quality, error rates, exception rates, user effort, control performance, and decision outcomes.

Where feasible, pilots should incorporate comparison groups or other credible methods of establishing whether observed improvements are actually attributable to AI. Not every finance process will permit a formal experimental design, but the underlying principle remains important: organizations should distinguish genuine performance improvements from changes caused by other factors.

The pilot should also explicitly evaluate human-AI interaction. A system that produces rapid outputs may still create limited value if employees spend substantial time checking or correcting those outputs. Conversely, an AI system that does not eliminate much manual work may nevertheless generate significant value if it enables professionals to focus on more complex and strategically important activities.

Governance should be embedded within the experiment rather than added after the pilot has demonstrated technical success. Data protection, security, model validation, explainability, accountability, and escalation requirements should be tested alongside productivity and quality. Almeida and Santos Júnior (2025) and Hadley et al. (2025) demonstrate the importance of organizational structures and capabilities in responsible AI governance. A pilot should therefore provide evidence not only about whether the technology works but also about whether the organization can govern it effectively.

The outcome of Stage 3 should be a decision based on evidence: scale, redesign, modify, pause, or terminate. A successful pilot is one that demonstrates sufficient value, reliability, organizational fit, and governance readiness to justify further investment.

10.4 Stage 4: Redesign — Reconfigure the Finance Process

The fourth stage is process redesign. Successful pilots should not simply be scaled as additional software layered onto existing workflows. The underlying finance process should be reconsidered in light of the capabilities that AI introduces.

This distinction is critical. If AI is added to an unchanged process, the organization may achieve incremental productivity improvements while retaining unnecessary handoffs, duplicated activities, manual controls, and legacy assumptions about how work should be performed. Greater value may instead be achieved by redesigning the process around the new human-AI configuration.

For example, a conventional reporting process might involve manual data preparation, spreadsheet analysis, management review, and report production. An AI-enabled process could instead involve automated data ingestion, AI-supported analysis, anomaly identification, human validation, management decision-making, and continuous feedback into the process.

The important transformation is not the introduction of the AI system itself. It is the reallocation of activities and decision rights across the process.

This principle is consistent with Abbas (2026), who emphasizes that AI can affect organizational structures, information flows, professional roles, and management-accounting practices. AI implementation should therefore be accompanied by explicit decisions concerning which tasks are automated, which are augmented, which remain human responsibilities, and where human approval is required.

Human-AI collaboration should be designed rather than left to informal employee behaviour. Dell'Acqua et al. (2025) demonstrate that AI performance depends on the relationship between the technology and the task. Employees must therefore know when AI can be relied upon, when its outputs require additional validation, and when a task lies outside the system's reliable capability range.

This is also where the potential risks of deskilling identified in Abbas (2026) become particularly relevant. Process redesign should ensure that employees retain sufficient expertise to understand and challenge AI outputs. Automation should reduce unnecessary work without eliminating the professional capabilities required to supervise the process effectively.

Process redesign should therefore address at least four questions. First, what should the AI system do? Second, what should humans do? Third, where should control and approval points be located? Fourth, how should information and exceptions flow through the redesigned process?

The answers should be determined by the characteristics and risks of the particular process rather than by a general objective of maximizing automation.

10.5 Stage 5: Institutionalize — Build AI into the Finance Operating Model

The fifth stage is institutionalization. Once AI-enabled processes have demonstrated value and have been appropriately redesigned, AI should become part of the finance operating model rather than remain a collection of isolated projects.

Institutionalization requires portfolio management, governance, continuous monitoring, model validation, workforce development, performance measurement, and periodic reassessment of AI applications. The organization must be capable of determining whether systems continue to generate value, whether risks have changed, whether processes remain appropriate, and whether new technologies provide better alternatives.

At this stage, governance becomes particularly important. Almeida and Santos Júnior (2025) and Hadley et al. (2025) indicate that responsible AI requires organizational structures rather than isolated technical controls. Finance organizations should therefore establish repeatable mechanisms for approving, monitoring, reviewing, and, where necessary, retiring AI applications.

Institutionalization should also include continuous performance measurement. The baseline established during the experimentation stage should not disappear once a project is scaled. Instead, it should become part of an ongoing performance-management framework. This allows the organization to determine whether expected benefits are being realized and whether new costs, risks, or unintended consequences have emerged.

The workforce dimension must likewise continue beyond implementation. As AI capabilities evolve, finance professionals require ongoing development in AI literacy, data interpretation, process design, professional judgment, and governance. The transformation of the finance profession described by Abbas (2026) should therefore be treated as a continuing organizational process rather than a one-time training programme.

Portfolio management remains important at the institutional stage. The organization should maintain a balanced set of AI investments, ranging from relatively low-risk productivity applications to more ambitious process transformations and selective strategic experiments. This approach reduces dependence on individual projects and creates a mechanism for allocating resources according to demonstrated performance and strategic priorities.

Institutionalization should also include periodic reassessment. An AI application that is valuable today may become less attractive as technology, costs, regulations, data environments, or organizational priorities change. Conversely, applications that were previously infeasible may become viable as systems improve and organizational capabilities mature. The AI portfolio should therefore be treated as dynamic rather than permanent.

10.6 The Five Stages as a Continuous Transformation Cycle

Although the five stages are presented sequentially, they should not be understood as a strictly linear process. Institutionalized AI systems generate new data, experience, and organizational learning that should feed back into diagnosis and prioritization. Similarly, the discovery of new risks during implementation may require organizations to revisit earlier assumptions about feasibility or governance.

The model is therefore better understood as a continuous transformation cycle. Diagnosis identifies opportunities and constraints; prioritization allocates scarce organizational resources; experimentation generates evidence; redesign converts successful experiments into improved processes; and institutionalization embeds the resulting capabilities into the finance operating model. Experience from institutionalized applications then informs the next cycle of diagnosis and prioritization.

This cyclical structure addresses several of the principal implementation risks identified by the Parnova framework. It reduces pilot proliferation by requiring prioritization, limits technology-driven experimentation by requiring measurable baselines, incorporates governance before scaling, and addresses organizational change through explicit process and workforce redesign (Parnova Consulting, 2026).

The model also integrates the major insights of the academic literature. The evidence from Noy and Zhang (2023), Brynjolfsson et al. (2025), and Dell'Acqua et al. (2025) supports the need for task-specific experimentation and rigorous measurement. Abbas (2026), Dong et al. (2024), and Stratopoulos and Wang (2025) highlight the broader transformation of accounting work, professional roles, and organizational capabilities. Eisikovits et al. (2024) emphasize the continuing importance of professional judgment, trust, and governance, while Almeida and Santos Júnior (2025) and Hadley et al. (2025) demonstrate the organizational foundations required for responsible AI governance.

The resulting model therefore shifts the focus of CFO AI strategy from technology acquisition to organizational capability building. The relevant question at each stage is not simply whether a new AI tool is available, but whether the organization can use that capability to improve a specific finance process while maintaining appropriate standards of performance, control, accountability, and professional judgment.

The ultimate objective is consequently not to create an “AI project” within finance. It is to create an AI-enabled finance function in which technology, processes, people, data, governance, and decision rights are deliberately integrated.

In this sense, the five-stage model can be summarized conceptually as follows: diagnose the organization before selecting technology; prioritize opportunities according to value and readiness; experiment to establish evidence; redesign processes around successful human-AI configurations; and institutionalize AI as a governed, measurable, continuously developing finance capability.

This approach provides a more robust foundation for CFO-led AI transformation than either technology-first implementation or unrestricted experimentation. It recognizes that the central challenge is not acquiring AI capability but converting that capability into sustainable organizational value.

11. Implications for CFOs and Boards

The preceding analysis has several implications for CFOs and boards. The central implication is that AI should no longer be regarded primarily as a technology investment or an operational efficiency programme. Its adoption increasingly represents a question of capital allocation, organizational design, risk governance, and the future role of the finance function.

11.1 AI as a Capital-Allocation Decision

First, AI investment should be treated explicitly as a capital-allocation question. CFOs should evaluate AI initiatives against alternative uses of organizational resources, including investments in enterprise systems, process redesign, analytics, workforce development, conventional automation, and additional human capacity.

This perspective is important because the availability of an attractive AI application does not establish that it is the best investment. A finance function may be able to automate a process using generative AI, for example, but process standardization or conventional workflow automation may deliver a more reliable return at lower cost and risk. Conversely, an AI investment may appear relatively expensive when measured only through labour savings but generate substantial additional value through improved forecasting, faster decision-making, or stronger risk detection.

The empirical literature reinforces the need for this context-specific assessment. Noy and Zhang (2023), Brynjolfsson et al. (2025), and Dell'Acqua et al. (2025) all demonstrate that AI effects vary according to tasks and users. The implication for CFOs is that generalized productivity claims should not substitute for organization-specific business cases.

AI investment should therefore be evaluated according to the specific economic problem being addressed, the expected source of value, the organizational capabilities required, and the risks introduced by the technology. The relevant question is not whether AI is valuable in general, but whether a particular AI-enabled intervention represents the best available use of organizational resources.

11.2 ROI Should Include Decision Quality

Second, AI business cases should extend beyond direct labour savings. Reducing manual effort is an important source of value for many finance applications, particularly in transaction processing, reporting, reconciliation, and document-intensive activities. However, some of the most strategically important applications generate value through improved decision quality rather than through simple reductions in employee hours.

AI-supported forecasting, liquidity management, fraud detection, anomaly identification, scenario analysis, and margin management may affect financial outcomes without producing proportionate reductions in headcount. Improved forecasts may allow earlier corrective action. Better anomaly detection may reduce losses. Faster scenario analysis may improve strategic responsiveness. More timely management information may shorten decision cycles.

This distinction is consistent with Abbas (2026), who emphasizes that AI affects management accounting information, organizational structures, professional roles, and decision processes. The value of AI should therefore be assessed at the level of the finance process and resulting organizational decisions, rather than solely at the level of the technology.

For boards and CFOs, this means that AI investment cases should identify both direct and indirect value mechanisms. Relevant evidence may include improvements in forecast quality, working-capital outcomes, control effectiveness, loss avoidance, decision speed, planning-cycle duration, and the ability of finance professionals to devote more time to strategically important activities.

At the same time, decision-quality benefits should be evaluated carefully. AI-generated recommendations do not automatically improve decisions. The research of Dell'Acqua et al. (2025) demonstrates that AI can improve performance within some tasks while potentially worsening it outside its effective capability frontier. Consequently, the business case must account for the possibility of inappropriate reliance on AI and the costs associated with validation, exceptions, and errors.

11.3 AI Maturity Depends on Process Maturity

Third, AI maturity depends substantially on process maturity. AI cannot reliably compensate for fragmented processes, inconsistent definitions, poor master data, unclear ownership, or ambiguous decision rights.

This point is central to the CFO AI Navigator 2026, which emphasizes organizational readiness, data quality, process assessment, and focused use-case selection before implementation (Parnova Consulting, 2026). The academic literature provides complementary support. Abbas (2026) shows that AI adoption affects organizational structures and information flows, while the governance literature emphasizes the importance of organizational capabilities and processes for responsible AI implementation.

In practice, AI frequently functions as an organizational stress test. The attempt to automate a finance process can expose weaknesses that previously remained hidden. An organization may discover that different business units use inconsistent definitions, that master data are unreliable, that approval responsibilities are unclear, or that employees rely on undocumented workarounds.

These problems are not necessarily caused by AI. Rather, AI makes them more visible because effective automation requires greater consistency and transparency in the underlying process. A finance organization may therefore need to standardize and simplify a process before introducing advanced AI.

This has an important strategic implication. AI transformation and finance-process transformation should not be treated as completely separate agendas. In some cases, the most valuable first step toward AI adoption may be process simplification, data remediation, or clarification of decision rights rather than immediate deployment of an AI system.

11.4 Governance as an Enabler of Scale

Fourth, governance should be understood as an enabler of responsible scaling rather than merely as a constraint on innovation.

Organizations sometimes perceive governance requirements as barriers that slow AI adoption. However, the absence of appropriate governance can create an even greater obstacle: executives may be unwilling to scale applications whose operational, legal, security, or reputational risks they cannot adequately understand or control.

The research of Almeida and Santos Júnior (2025) demonstrates that effective AI governance depends on organizational structures and capabilities, while Hadley et al. (2025) highlight the importance of leadership support and integration with existing organizational processes. These findings suggest that governance should be embedded within the operating model of AI rather than introduced as a final approval step.

For finance organizations, effective governance provides the basis for determining which AI applications can be used, under what conditions, with what degree of autonomy, and subject to which controls. It can establish clear responsibilities for data ownership, model validation, human review, security, monitoring, escalation, and accountability.

This is particularly important as organizations move from relatively low-risk generative AI applications toward systems capable of influencing or executing financial processes. The greater the potential consequence of an AI-supported action, the more important it becomes to establish explicit decision rights and oversight mechanisms.

Good governance can therefore accelerate responsible adoption. When employees and executives understand the boundaries within which AI can operate, experimentation becomes easier to manage and successful applications become easier to scale. Governance creates organizational confidence by making risk visible and controllable.

11.5 The CFO as the Integrator of Technology, Economics, and Risk

Fifth, the CFO role itself is evolving. The future CFO increasingly needs to understand not only financial performance but also how technology changes the organization's economics, controls, workforce, processes, and decision architecture.

This does not mean that CFOs need to become AI engineers. Rather, they need sufficient technological and organizational understanding to ask the right strategic questions: Where can AI create material value? Which processes are suitable for automation or augmentation? What data and systems are required? Where should human judgment remain central? What risks are being introduced? How should performance be measured? And what organizational capabilities are required for scale?

The CFO is particularly well positioned to integrate these considerations because the finance function sits at the intersection of resource allocation, performance measurement, risk management, internal control, forecasting, and strategic decision-making. The CFO AI Navigator 2026 therefore appropriately positions CFO ownership as a central condition for successful finance AI transformation (Parnova Consulting, 2026).

This leadership responsibility also changes the relationship between finance and IT. AI transformation should not be separated into a “business” side that identifies use cases and a “technology” side that implements them. Effective transformation requires joint ownership. IT and data functions remain essential for architecture, cybersecurity, integration, technical implementation, and vendor management, while finance leadership must determine business priorities, materiality, process design, control requirements, and economic value.

The CFO therefore becomes an integrator rather than simply a technology sponsor.

11.6 Implications for Boards

These developments also have implications for boards. Boards should increasingly treat AI as an element of corporate strategy and organizational risk rather than as a technical issue delegated entirely to management or IT.

Board oversight should focus on whether management has established a coherent AI strategy, whether investments are aligned with corporate priorities, whether expected benefits are being measured, and whether material risks are appropriately governed. The board should also understand where AI is being used in financially significant processes and what degree of autonomy has been granted to AI systems.

Particular attention should be given to applications affecting financial reporting, internal controls, fraud detection, treasury, forecasting, customer or employee data, and other areas where AI errors could have material consequences.

Boards should also consider whether the organization possesses the capabilities necessary to govern AI effectively. This includes appropriate expertise among management, clear accountability, effective internal controls, adequate data governance, cybersecurity capabilities, and mechanisms for independent challenge.

The increasing importance of AI also creates a potential board-level skills question. Just as boards have developed expectations concerning financial literacy and cybersecurity oversight, they may increasingly need sufficient AI literacy to challenge management assumptions and understand the strategic and risk implications of AI investments.

11.7 From Technology Strategy to Decision Architecture

The broader implication is that AI strategy should ultimately be connected to decision architecture. AI changes not only how information is processed but also how decisions are prepared, challenged, approved, and executed.

This is particularly important for finance because the function's core contribution is not simply the production of financial information. Finance supports decisions concerning resource allocation, performance, investment, liquidity, risk, and strategy. If AI changes the information available to decision-makers or the speed at which it is produced, it can consequently change the structure of managerial decision-making itself.

Stratopoulos and Wang (2025) emphasize the broader transformation that AI is producing within accounting research and practice. Dong et al. (2024) similarly demonstrate the breadth of emerging LLM applications across accounting and finance. These developments suggest that CFOs should think beyond individual tools and instead consider how AI changes the architecture through which the finance function creates information and supports decisions.

This also reinforces the importance of human judgment. Eisikovits et al. (2024) highlight the continuing risks associated with trust, governance, bias, and professional judgment in AI-enabled accounting and auditing. As AI becomes more capable, the responsibility of finance professionals may shift increasingly toward interpreting AI outputs, challenging assumptions, assessing uncertainty, and ensuring that decisions remain appropriately accountable.

11.8 Strategic Implication

Taken together, these implications suggest that the central challenge for CFOs is not determining whether AI should be adopted. For most finance organizations, the more important question is where, how, and under what conditions AI should be integrated into the finance operating model.

The evidence reviewed in this paper points toward a disciplined answer. CFOs should treat AI as a capital-allocation decision; evaluate both efficiency and decision-quality benefits; strengthen process and data foundations; establish governance as an organizational capability; invest in workforce transformation; and retain appropriate human accountability for material decisions.

This perspective also changes how AI maturity should be understood. An organization should not be considered AI-mature simply because it has deployed many AI applications or uses sophisticated models. A more meaningful indicator of maturity is the organization's ability to identify valuable opportunities, evaluate them empirically, govern them appropriately, redesign processes around them, and scale successful applications while maintaining professional judgment and organizational accountability.

The ultimate responsibility of the CFO is therefore not to maximize AI adoption. It is to maximize the sustainable value created by AI within the finance function and the wider organization.

In this respect, the CFO's emerging role is broader than technology sponsorship. It involves integrating technology, economics, process design, governance, organizational capability, and strategic decision-making. The finance leaders most likely to capture lasting value from AI will be those who recognize that the fundamental transformation is not from manual work to automated work, but from a traditional finance operating model to an AI-enabled, human-governed decision system.

12. Research Propositions and Future Research Agenda

aThe combination of the practitioner perspective developed in the CFO AI Navigator 2026 and the emerging academic literature suggests a research agenda that moves beyond questions of whether artificial intelligence can perform accounting and finance tasks. The more consequential questions concern under what organizational conditions AI creates sustainable value, how that value should be measured, and how finance organizations should govern the changing relationship between human expertise and AI capabilities.

This shift is important because the current literature remains considerably more developed in demonstrating the technical potential and short-term productivity effects of AI than in explaining the organizational mechanisms through which those effects translate into sustained financial performance. Abbas (2026) identifies precisely this need in the management-accounting literature, highlighting unresolved questions concerning AI adoption, organizational change, professional judgment, trust, explainability, and the consequences of emerging technologies. Stratopoulos and Wang (2025) similarly identify a broad research agenda for accounting and AI, while Dong, Stratopoulos, and Wang (2024) demonstrate the rapid expansion of research on generative AI and large language models across accounting and finance.

Against this background, six propositions provide a foundation for future empirical research.

12.1 Proposition 1: CFO Ownership Increases the Probability of Successful Finance-AI Scaling

The first proposition is that CFO ownership increases the probability that successful AI experiments will progress to sustainable organizational adoption.

The rationale follows from the distinctive position of the finance function. AI implementation in finance involves more than technology selection. It requires decisions about process economics, internal controls, materiality, resource allocation, risk tolerance, workforce design, and the measurement of financial outcomes. These issues fall directly within the CFO's responsibilities.

The Parnova framework therefore argues that AI should be treated as a strategic finance issue rather than as an IT initiative (Parnova Consulting, 2026). The academic literature provides a complementary basis for examining this proposition. Almeida and Santos Júnior (2025) demonstrate that effective AI governance involves multiple organizational layers, while Hadley et al. (2025) emphasize the importance of leadership support and integration with existing organizational processes.

Future research could test whether AI initiatives with direct CFO sponsorship or joint business-technology governance are more likely to progress beyond experimentation than initiatives primarily managed as technology projects. Such research could compare organizations according to governance structure, executive sponsorship, decision rights, and the extent of finance leadership involvement.

Longitudinal research would be particularly valuable because the effect of leadership may not be visible during initial experimentation. CFO ownership may matter most during the transition from a technically successful pilot to enterprise-wide implementation, when questions of funding, process redesign, controls, organizational change, and workforce development become more consequential.

12.2 Proposition 2: AI Value Is Greater When Use-Case Selection Is Based on Process Economics Rather Than Technological Novelty

The second proposition is that AI investments selected according to process economics and strategic relevance generate greater realized value than investments selected primarily according to technological novelty.

The rapid development of generative and agentic AI creates a substantial risk of technology-driven experimentation. Organizations may pursue highly visible applications because the underlying technology is novel rather than because the application addresses a material business problem.

The Parnova framework proposes the opposite approach: organizations should assess current processes, identify high-value opportunities, establish baselines, and prioritize a limited number of use cases according to expected value and feasibility (Parnova Consulting, 2026).

This proposition could be investigated through longitudinal studies of AI portfolios. Researchers could examine whether organizations that systematically evaluate transaction volumes, manual effort, process bottlenecks, data readiness, risk, and expected business value achieve higher realized returns than organizations whose adoption decisions are driven primarily by technological opportunity.

Such research would also help address an important limitation in the current literature. Noy and Zhang (2023), Brynjolfsson et al. (2025), and Dell'Acqua et al. (2025) provide strong evidence that AI can affect individual or task-level productivity, but these findings do not by themselves establish which organizational investments produce the greatest economic returns. The missing analytical step is the connection between task-level AI performance and portfolio-level capital allocation.

Future research should therefore examine not only whether AI improves a task but also whether organizations are selecting the right tasks in which to invest.

12.3 Proposition 3: Data and Process Maturity Moderate the Relationship Between AI Adoption and Performance

The third proposition is that the performance effects of AI adoption are stronger in organizations with mature, standardized, and integrated finance processes.

AI systems operate within organizational environments. Their effectiveness therefore depends partly on the quality, accessibility, consistency, and governance of the underlying data and processes. Fragmented systems, inconsistent definitions, unreliable master data, and unclear process ownership may reduce the value of otherwise capable AI systems.

This proposition extends the Parnova emphasis on organizational readiness and provides an important direction for empirical research. Instead of treating AI adoption as an independent factor that produces a uniform performance effect, researchers should examine the organizational conditions under which adoption occurs.

For example, two companies implementing similar AI-enabled forecasting systems may achieve very different results because one has integrated data and standardized planning processes while the other relies on fragmented spreadsheets and inconsistent business-unit definitions.

Future research could therefore examine data quality, process standardization, system integration, and governance maturity as moderating variables in the relationship between AI adoption and finance-function performance. Such research could provide CFOs with more realistic expectations concerning the organizational prerequisites for AI value creation.

This perspective also connects with Abbas (2026), whose review emphasizes that AI affects organizational structures and information flows rather than operating solely as an isolated technological intervention. AI implementation may consequently create greater value when embedded within mature organizational systems.

12.4 Proposition 4: Human-AI Complementarity Produces Greater Sustainable Value Than Unrestricted Automation in High-Judgment Finance Activities

The fourth proposition is that human-AI complementarity generates greater sustainable value than unrestricted automation for finance activities characterized by substantial judgment, uncertainty, and contextual interpretation.

The distinction between automation and augmentation is increasingly important. AI can perform information-intensive activities such as classification, summarization, anomaly detection, forecasting, and document analysis. However, financial decisions frequently involve ambiguity, materiality judgments, incomplete information, and contextual considerations that cannot be reduced to pattern recognition alone.

The empirical evidence provides a strong foundation for investigating this proposition. Brynjolfsson et al. (2025) find substantial productivity benefits from generative AI while also demonstrating heterogeneity across workers. Dell'Acqua et al. (2025) show that AI performance depends on whether tasks fall within the system's effective capability frontier. Together, these findings challenge the assumption that more automation necessarily produces better outcomes.

Accounting research raises similar concerns. Eisikovits et al. (2024) emphasize risks involving professional judgment, trust, governance, and potential deskilling, while Abbas (2026) identifies the continuing importance of human judgment alongside the increasing automation of routine activities.

Future research could compare three organizational configurations: predominantly automated decision-making, human-in-the-loop decision-making, and predominantly human-led decision-making supported by AI. Outcomes could include decision quality, productivity, error rates, professional learning, employee confidence, control effectiveness, and long-term organizational resilience.

Such research would be particularly valuable in areas such as financial forecasting, accounting judgments, fraud investigation, liquidity decisions, internal audit, and management reporting, where the cost of an incorrect AI-supported decision may substantially exceed the productivity benefit of automation.

12.5 Proposition 5: Governance Maturity Moderates the Scalability of AI

The fifth proposition is that AI governance maturity has a limited effect on low-risk experimentation but becomes increasingly important as AI applications move toward material decisions and autonomous execution.

This proposition follows directly from the distinction between experimentation and institutionalization. An employee using generative AI to summarize a non-material internal document presents a different governance challenge from an AI system that influences financial reporting, approves transactions, identifies fraud, or autonomously executes a sequence of financial activities.

The research of Almeida and Santos Júnior (2025) demonstrates that AI governance requires organizational structures and capabilities, while Hadley et al. (2025) highlight the importance of leadership support and integration with established organizational processes. These findings suggest that governance should be investigated not merely as a compliance mechanism but as a potential determinant of organizational scalability.

Future empirical studies could examine whether organizations with more mature governance systems are able to scale AI applications more rapidly and safely than organizations with weak or fragmented governance. Relevant governance dimensions could include accountability structures, model validation, data governance, monitoring, human oversight, escalation mechanisms, security controls, and employee training.

An especially important research question concerns the relationship between governance and innovation. Governance may initially appear to constrain experimentation, but sufficiently mature governance could instead facilitate innovation by creating clear boundaries within which experimentation can occur. Research should therefore distinguish between bureaucratic control and enabling governance.

12.6 Proposition 6: AI Shifts Finance Performance from Information Production Toward Decision Enablement

The sixth proposition is that as AI reduces the cost and time required to produce financial information, the relative strategic importance of interpretation, scenario analysis, business partnering, and decision support increases.

This proposition represents perhaps the most consequential long-term implication for the finance function.

Traditional finance performance has often been evaluated through the reliability and timeliness of information production: closing accounts, preparing reports, reconciling transactions, producing forecasts, and maintaining financial controls. AI has the potential to automate or accelerate many of these activities.

As the cost of producing information falls, however, information itself may become less scarce. The scarce organizational resource may increasingly become the ability to interpret information, distinguish relevant signals from noise, evaluate alternative scenarios, understand uncertainty, and translate analysis into effective managerial action.

This proposition is consistent with the broader implications identified by Abbas (2026), Dong et al. (2024), and Stratopoulos and Wang (2025), all of whom highlight the implications of AI for accounting roles, information processes, and professional capabilities.

It also aligns with the productivity evidence from Noy and Zhang (2023) and Brynjolfsson et al. (2025). If AI reduces the time required to complete routine knowledge-work activities, the resulting capacity does not automatically create organizational value. Organizations must decide how to redeploy that capacity. In finance, the opportunity may be to shift employee effort from information production toward higher-value activities such as business partnering, strategic analysis, risk assessment, and decision support.

Future research should therefore examine whether AI adoption changes the allocation of finance professionals' time and whether such changes are associated with improved organizational outcomes. Rather than measuring only reductions in processing time, researchers could investigate changes in the proportion of finance work devoted to transactional activities, analytical activities, strategic decision support, and governance.

12.7 Toward a More Mature Empirical Research Agenda

These propositions point to a broader limitation in the current literature. Much of the emerging evidence establishes what AI systems can do, but considerably less evidence establishes what organizations should do to capture and sustain that value.

This distinction is particularly important because technological capability and organizational performance are not equivalent. An AI model may demonstrate high technical accuracy without improving business outcomes. A productivity improvement at the task level may not translate into organizational productivity if employees spend additional time validating outputs or resolving exceptions. Similarly, successful experimentation does not necessarily lead to successful enterprise-scale adoption.

Abbas (2026) explicitly identifies the need for additional empirical research into the organizational, professional, and managerial consequences of AI in management accounting. Stratopoulos and Wang (2025) likewise identify significant opportunities for research examining how AI changes accounting practices and research methods. Dong et al. (2024) demonstrate the rapid expansion of research into LLM applications while also revealing how early the empirical literature remains in many areas.

Future research should consequently place greater emphasis on longitudinal field studies, controlled field experiments, comparative case studies, and large-sample quantitative research. These approaches can help distinguish temporary productivity effects from sustainable organizational transformation.

Longitudinal research is particularly important because AI adoption is dynamic. Initial productivity gains may change as employees learn to use AI, processes are redesigned, systems mature, and organizations develop new governance mechanisms. Conversely, initial gains may disappear if systems are poorly maintained, data quality deteriorates, or employees become excessively dependent on AI.

Field experiments can provide stronger causal evidence concerning the effects of different implementation models. For example, researchers could compare finance teams using AI with and without structured training, baseline measurement, human review, or governance mechanisms. Comparative case studies could then explain why similar technologies generate different outcomes across organizations.

Another important area concerns measurement. Future research should distinguish technical performance from organizational value. Model accuracy, response time, or task completion time are useful indicators, but they should be complemented by measures of decision quality, financial outcomes, control effectiveness, employee capability, risk, and long-term organizational performance.

Research should also examine potential negative outcomes more systematically. These include deskilling, automation bias, excessive reliance on AI, loss of professional expertise, hidden validation costs, cybersecurity risks, privacy failures, and the concentration of decision-making power in technological systems. Eisikovits et al. (2024) provide an important starting point for this line of inquiry.

Finally, research should examine the evolution from assistive AI toward agentic AI. Dong et al. (2024) document the rapid development of LLM applications, while the broader trajectory of AI suggests increasing movement from systems that generate recommendations toward systems capable of coordinating and executing multiple activities. This transition could fundamentally alter the control environment of finance. The research question is therefore not simply whether autonomous AI can perform finance tasks, but how organizations should design accountability, human oversight, controls, and decision rights when AI systems increasingly participate directly in financial processes.

12.8 Concluding Research Direction

Taken together, the six propositions suggest a shift in the research agenda from AI capability to AI-enabled organizational performance.

The next generation of research should investigate the conditions under which AI creates durable value, rather than treating adoption itself as the principal outcome. It should examine the interaction between technology and organizational context, including leadership, process maturity, data quality, governance, workforce capabilities, and human judgment.

Such an agenda would also strengthen the connection between academic research and the practical challenges identified by Parnova Consulting (2026). The CFO AI Navigator 2026 emphasizes CFO ownership, focused use-case selection, baseline measurement, governance, change management, and scalable process design. The academic literature provides theoretical and empirical foundations for many of these principles, but further research is required to determine their relative importance and causal effects.

The central research challenge can therefore be expressed more precisely: how do organizations convert AI capabilities into measurable, governable, and sustainable improvements in finance-function performance?

Answering this question requires moving beyond demonstrations of what AI can accomplish and toward systematic investigation of how technology, people, processes, governance, and organizational decision-making interact. This is where the emerging research agenda in accounting and finance can make its most significant contribution.

13. Discussion

aThe CFO AI Navigator 2026 is most valuable when interpreted as a managerial implementation framework rather than as an empirical study of artificial intelligence. Its central proposition—that AI adoption in finance is fundamentally a leadership and organizational challenge rather than simply an IT initiative—is strongly supported by the emerging academic literature. At the same time, the academic evidence provides several important qualifications that sharpen and extend the Parnova perspective.

13.1 From Technological Potential to Context-Dependent Value

The first refinement concerns the assumption that AI generates uniform productivity benefits. The evidence increasingly demonstrates that AI's effects are highly dependent on the task, user, organizational context, and manner of implementation.

Noy and Zhang (2023) provide experimental evidence that generative AI can substantially reduce task completion time while improving output quality. Brynjolfsson, Li, and Raymond (2025) similarly find significant productivity improvements among customer-support workers, but their results also demonstrate substantial variation across workers. Dell'Acqua et al. (2025) provide perhaps the clearest qualification: AI can improve performance when tasks fall within its effective capability frontier but may reduce performance when users apply it to tasks beyond that frontier.

These findings have important implications for the interpretation of AI business cases. Statements such as “AI increases productivity by a particular percentage” should not be treated as universal parameters. Productivity effects observed in one occupation, organization, or task environment cannot automatically be transferred to another.

For CFOs, the relevant question is therefore not whether AI has a demonstrated average productivity effect, but where within the organization's finance processes AI is likely to produce a positive and reliable effect.

This reinforces the Parnova emphasis on use-case selection and baseline measurement. AI investments should be evaluated at the process level, with attention to task characteristics, employee capabilities, data quality, process maturity, and the extent to which AI capabilities match the requirements of the activity.

13.2 From Generic ROI Claims to Contextual Business Cases

The second refinement concerns the interpretation of reported return-on-investment figures. The Parnova guide provides useful practitioner benchmarks and ROI signals for applications such as invoice processing, financial close, expense management, and forecasting (Parnova Consulting, 2026). Such estimates are valuable for identifying potentially attractive opportunities, but they should not be interpreted as universal empirical benchmarks.

The economic value of an AI application depends on the organization's starting point. A company with highly manual invoice processing may have substantial scope for automation, while an organization with a mature, highly automated accounts-payable environment may have considerably less incremental value available.

The same principle applies to financial close, reporting, forecasting, and other applications. The value generated by AI depends on existing technology, transaction volumes, process design, data quality, employee effort, error rates, and the cost of implementation and governance.

This contextual perspective is consistent with the broader academic literature. The productivity studies of Noy and Zhang (2023), Brynjolfsson et al. (2025), and Dell'Acqua et al. (2025) demonstrate that AI outcomes depend on the relationship between technology and the specific work being performed. Abbas (2026) extends this logic to management accounting by emphasizing the organizational and professional consequences of AI adoption.

Consequently, a robust CFO business case should begin with the economics of the existing process, rather than with the capabilities of the proposed AI technology. The question should be what value remains to be captured, what constraints prevent that value from being captured today, and whether AI represents the most effective intervention.

13.3 Governance as an Organizational Capability

The third refinement concerns AI governance. The Parnova framework appropriately identifies privacy, model validation, explainability, and governance as essential components of AI implementation. The academic literature suggests that these should be understood not simply as compliance requirements but as organizational capabilities that enable responsible scaling.

Almeida and Santos Júnior (2025) demonstrate that AI governance involves multiple organizational layers and that training and organizational capabilities are associated with more advanced governance practices. Hadley, Blatecky, and Comfort (2025) similarly show the importance of leadership support and integration with existing organizational processes in the operation of responsible-AI governance mechanisms.

This evidence challenges the view that governance can be reduced to the creation of an AI policy or a legal review process. Effective governance requires clear accountability, appropriate expertise, organizational processes, monitoring mechanisms, and the ability to integrate AI oversight into existing management structures.

The distinction becomes increasingly important as finance organizations move from assistive applications toward more autonomous systems. A generative AI tool used to prepare an internal draft presents a fundamentally different governance challenge from an AI system that recommends accounting treatments, influences liquidity decisions, identifies potentially fraudulent transactions, or executes multiple financial activities.

Governance requirements should therefore be proportionate to the consequences of AI-supported decisions. The greater the materiality, autonomy, and potential impact of an AI application, the greater the need for validation, human oversight, monitoring, documentation, and accountability.

This perspective also explains why governance should be established early rather than after successful experimentation. Without governance mechanisms, organizations may discover that technically successful pilots cannot be scaled because executives are unwilling to accept the associated operational, regulatory, security, or reputational risks.

13.4 Human-AI Complementarity and the Future of Finance Work

The fourth refinement concerns the nature of transformation itself. The evidence does not support a simple narrative in which AI either leaves work unchanged or replaces human professionals. A more plausible trajectory is one of progressive reconfiguration of work.

Routine activities are likely to become increasingly automated. Analytical activities are likely to become increasingly augmented. Some processes may eventually become substantially autonomous. Yet professional judgment, contextual interpretation, accountability, and business communication remain important, particularly in high-stakes financial decisions.

The research of Brynjolfsson et al. (2025) demonstrates the importance of worker experience and the interaction between AI and human capabilities. Dell'Acqua et al. (2025) further show that AI effectiveness depends on task characteristics. These findings support an operating model in which AI capabilities are deliberately matched to tasks and combined with human expertise rather than deployed according to a blanket objective of maximum automation.

Accounting research reinforces this interpretation. Abbas (2026) identifies both opportunities for automation and the continuing importance of human judgment, while Eisikovits et al. (2024) highlight risks involving trust, bias, governance, and professional capabilities.

The implication is that the future finance function will likely contain a spectrum of human-AI configurations. Some activities will be highly automated, some will be AI-assisted, and others will remain predominantly human-led. The appropriate configuration will depend on the complexity of the task, the consequences of error, the availability of reliable data, and the importance of contextual judgment.

13.5 From Task Automation to Process Transformation

A further implication is that the most important unit of analysis should be the finance process rather than the AI application.

This distinction is central to understanding why some AI implementations produce limited value despite impressive technical capabilities. Deploying AI into an unchanged process may automate one activity while leaving inefficient handoffs, duplicated controls, fragmented data, or unnecessary approvals intact.

By contrast, AI can create greater value when it becomes part of a redesigned process in which activities, decision rights, controls, and human responsibilities are reconsidered simultaneously.

This proposition is consistent with Abbas's (2026) finding that AI affects organizational structures, information flows, professional roles, and management-accounting practices. It is also consistent with the Parnova framework's emphasis on process assessment, organizational change, and scalability (Parnova Consulting, 2026).

The distinction matters because the economic consequences of AI may extend well beyond direct labour savings. Process redesign can change cycle times, control structures, information availability, decision latency, and the allocation of managerial attention. These effects may ultimately be more consequential than the automation of individual tasks.

13.6 From Information Production to Decision Enablement

The discussion also suggests a broader transformation in the role of finance.

Historically, a substantial proportion of finance effort has been devoted to producing reliable information: collecting data, reconciling accounts, preparing reports, constructing forecasts, and explaining variances. AI has the potential to reduce the time and effort required for many of these activities.

As information production becomes faster and less costly, however, the strategic value of finance may increasingly shift toward decision enablement. Finance professionals may spend relatively less time producing information and relatively more time interpreting it, evaluating scenarios, challenging assumptions, communicating uncertainty, and supporting strategic choices.

This interpretation is consistent with the literature reviewed by Abbas (2026), Dong et al. (2024), and Stratopoulos and Wang (2025), which collectively indicates that AI is affecting not only accounting tasks but also professional roles, organizational structures, and the broader information environment.

The transformation therefore should not be evaluated solely according to whether finance departments reduce the number of hours spent on reporting or transaction processing. A more strategically meaningful question is whether the organization is able to redeploy finance capacity toward activities that improve managerial decisions and organizational performance.

13.7 AI as an Organizational Stress Test

An additional insight emerging from the combined practitioner and academic perspectives is that AI can function as an organizational stress test. Rather than simply introducing a new technological capability, AI implementation can reveal the extent to which an organization’s existing data, processes, governance structures, and workforce capabilities are sufficiently mature to support technology-enabled transformation (Abbas, 2026; Almeida and Santos Júnior, 2025; Stratopoulos and Wang, 2025).

The attempt to introduce AI frequently exposes weaknesses in processes and data that existed before the technology was introduced. Inconsistent master data, fragmented systems, undocumented procedures, ambiguous ownership, and unclear decision rights can all become barriers to AI implementation. The importance of these organizational foundations is consistent with the AI governance literature, which emphasizes that effective AI implementation requires organizational structures capable of establishing accountability, oversight, and clearly defined responsibilities (Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025). Similarly, accounting research indicates that the value of AI depends on the quality of the information environment and the organizational context in which AI applications are deployed (Abbas, 2026; Stratopoulos and Wang, 2025).

This does not mean that AI necessarily causes these weaknesses. Rather, AI makes them more visible because effective AI-enabled processes require reliable information, clearly defined workflows, and explicit accountability. The technology therefore acts as a diagnostic mechanism: when an AI application requires standardized data, consistent processes, or clearly defined decision rules, deficiencies in these areas become more apparent. This is particularly relevant in finance, where AI applications may depend on information drawn from multiple systems and processes and where errors or inconsistencies can have material consequences (Abbas, 2026; Eisikovits, Johnson and Markelevich, 2024).

Consequently, unsuccessful AI implementation should not always be interpreted as evidence that the technology failed. In some cases, it reveals that the organization was not sufficiently prepared to use the technology effectively. The distinction between technological capability and organizational performance is important here. Evidence that AI produces different outcomes depending on task characteristics and organizational context suggests that the presence of a technically capable system does not guarantee improved performance (Noy and Zhang, 2023; Dell’Acqua et al., 2026). A failed pilot may therefore identify deficiencies in data, processes, skills, governance, or workflow design that need to be addressed before the organization can capture the intended benefits.

This provides an important qualification to the technology-centric view of digital transformation. AI readiness should be understood partly as a measure of organizational readiness. Research on AI governance supports this broader interpretation by emphasizing the importance of organizational structures, accountability mechanisms, review processes, and responsible implementation alongside technological capabilities (Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025). For finance functions, readiness therefore extends beyond access to AI tools to include the ability of employees to understand and challenge AI outputs, the maturity of underlying processes, and the existence of appropriate controls (Abbas, 2026; Eisikovits, Johnson and Markelevich, 2024).

Investments in data quality, process standardization, governance, and workforce capabilities may therefore be prerequisites for realizing the benefits of more advanced AI. This suggests that AI investment can have an indirect value even where an individual application does not immediately deliver the expected return: the implementation process may identify and motivate improvements in the organizational infrastructure required for future AI initiatives. In this respect, AI transformation can create a reinforcing cycle in which better data, more standardized processes, stronger governance, and greater workforce capability increase the organization’s capacity to adopt increasingly sophisticated AI applications (Stratopoulos and Wang, 2025; Abbas, 2026).

For CFOs, the practical implication is that AI readiness should be assessed before, and developed alongside, AI deployment. Rather than treating data remediation, process standardization, governance, and skills development as secondary activities that follow technology investment, these capabilities should be considered part of the investment required to make AI economically viable. The organization should therefore ask not only whether a particular AI solution is technically feasible, but also whether the surrounding finance environment is sufficiently mature to support reliable adoption and sustained value creation (Almeida and Santos Júnior, 2025; Parnova Consulting, 2026).

Viewed in this way, the organizational stress-test effect is not necessarily a disadvantage of AI adoption. It can represent a strategic opportunity. By exposing weaknesses that may previously have remained hidden within routine processes, AI implementation can provide finance leaders with a clearer understanding of where organizational foundations need to be strengthened. The resulting transformation is therefore broader than the deployment of AI itself: AI can become a catalyst for improving the data, processes, governance, and capabilities on which a digitally enabled finance function depends (Abbas, 2026; Stratopoulos and Wang, 2025).

13.8 Evolution Rather Than Technological Disruption Alone

The combined evidence also suggests that the transformation of finance is likely to be evolutionary rather than purely disruptive.

Generative AI is already changing how employees produce and interpret information. Traditional machine-learning applications continue to support prediction, anomaly detection, and classification. Conventional automation remains valuable for highly standardized activities. At the same time, agentic systems may increasingly coordinate multiple activities and introduce new forms of process autonomy.

These technologies are unlikely to affect all areas of finance simultaneously or in the same way. Instead, transformation will occur through different combinations of automation, augmentation, and autonomy.

Dong et al. (2024) demonstrate the breadth of emerging LLM applications across accounting and finance, while Stratopoulos and Wang (2025) highlight the rapidly developing research and professional implications of AI. Abbas (2026) similarly emphasizes that AI adoption affects skills, roles, structures, and professional boundaries.

The resulting finance function is therefore unlikely to be either fully automated or fundamentally unchanged. It is more likely to become a hybrid organizational system in which humans and AI perform complementary roles across different levels of the finance value chain.

13.9 A Refined Conceptual Argument

Taken together, these observations allow the central argument of this paper to be sharpened.

The strategic value of AI in finance does not primarily arise from the replacement of finance professionals or from the deployment of increasingly sophisticated models. It arises from reconfiguring the economic and organizational architecture through which finance work is performed. The emerging accounting literature similarly suggests that AI should be understood as a transformation of accounting activities, information processes, professional roles, and organizational decision-making rather than simply as a mechanism for automating individual tasks (Dong, Stratopoulos and Wang, 2024; Abbas, 2026; Stratopoulos and Wang, 2025).

This architecture includes the interaction among technology, data, processes, people, controls, governance, and decision rights. AI changes the capabilities available within this system, but the resulting value depends on how those capabilities are integrated into organizational processes. Research on AI governance demonstrates that technological implementation must be accompanied by appropriate accountability, oversight, and organizational structures if AI is to generate sustainable value (Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025). Similarly, evidence from knowledge-work settings indicates that AI performance depends on the characteristics of the task and the context in which the technology is used, rather than on model capability alone (Dell’Acqua et al., 2026).

The implication is important for both practice and research. For practitioners, it means that AI strategy should focus on process economics, organizational readiness, human-AI complementarity, governance, and measurable business outcomes. For researchers, it means that AI adoption should increasingly be studied as an organizational phenomenon rather than simply as a technological intervention. This broader perspective is consistent with the accounting research agenda, which identifies the organizational, behavioural, professional, and governance consequences of AI as important areas for further investigation (Dong, Stratopoulos and Wang, 2024; Stratopoulos and Wang, 2025).

The Parnova framework makes an important practical contribution by translating these principles into a CFO-oriented implementation approach: diagnose the current state, prioritize high-value opportunities, establish baselines, experiment, redesign processes, and institutionalize successful applications (Parnova Consulting, 2026). The academic literature strengthens this approach by demonstrating why task heterogeneity, human judgment, organizational capabilities, and governance matter. In particular, evidence concerning the uneven effects of AI across tasks suggests that experimentation and evidence-based evaluation are preferable to assuming that AI will produce uniform improvements across the finance function (Noy and Zhang, 2023; Brynjolfsson, Li and Raymond, 2025; Dell’Acqua et al., 2026).

The resulting synthesis can be expressed as a fundamental shift in perspective:

The question is not whether AI can perform finance activities. The strategic question is how finance organizations should redesign work, decision-making, governance, and professional capabilities so that AI produces sustainable organizational value.

This distinction separates AI adoption from AI transformation. Adoption concerns whether an organization uses AI. Transformation concerns whether AI changes how the organization creates value. The distinction is consistent with the broader literature, which suggests that the organizational consequences of AI depend on how technological capabilities are embedded within processes, roles, governance structures, and patterns of human decision-making (Almeida and Santos Júnior, 2025; Abbas, 2026; Stratopoulos and Wang, 2025).

For CFOs and boards, the latter is the strategically relevant objective. The ultimate measure of successful finance AI adoption is therefore not the number of tools deployed, models implemented, or pilots completed. It is whether the finance function becomes more productive, more analytically capable, more responsive, better governed, and more effective in supporting organizational decisions. This broader conception of performance is supported by evidence that AI can generate productivity improvements but that those improvements depend on the interaction between AI capabilities, task characteristics, worker experience, and organizational context (Noy and Zhang, 2023; Brynjolfsson, Li and Raymond, 2025; Dell’Acqua et al., 2026).

The strongest conclusion from the combined evidence is consequently not that AI will replace finance work. Rather, AI is likely to change what constitutes valuable finance work. Routine information production will increasingly be automated; analytical work will increasingly be augmented; governance will become more technologically sophisticated; and professional judgment, contextual interpretation, and strategic business partnering may become relatively more important (Abbas, 2026; Eisikovits, Johnson and Markelevich, 2024; Stratopoulos and Wang, 2025). This does not imply that human expertise becomes less important. Instead, the locus of professional value may shift from producing and processing information toward validating, interpreting, challenging, and acting upon increasingly AI-generated information.

The workforce implications reinforce this conclusion. Evidence that less-experienced workers can benefit substantially from generative AI assistance suggests that AI may alter traditional patterns of productivity and expertise within organizations (Brynjolfsson, Li and Raymond, 2025). At the same time, the limitations identified at the boundaries of AI capability mean that experienced professionals remain important for handling exceptions, evaluating uncertainty, applying contextual knowledge, and maintaining professional standards (Dell’Acqua et al., 2026; Abbas, 2026). The future finance function is therefore likely to depend on a combination of technological fluency, financial expertise, data literacy, professional judgment, and business-partnering capabilities.

Governance is similarly central to sustainable transformation. As AI becomes embedded in processes that influence financial reporting, forecasting, controls, and strategic decisions, organizations must establish appropriate mechanisms for accountability, monitoring, human oversight, and escalation (Eisikovits, Johnson and Markelevich, 2024; Almeida and Santos Júnior, 2025; Hadley, Blatecky and Comfort, 2025). Consequently, AI transformation cannot be separated from the development of the organizational capabilities required to govern its use responsibly.

The long-term competitive advantage will therefore belong not necessarily to organizations that adopt the most AI, but to those that integrate AI most effectively with human expertise, organizational processes, and disciplined governance. The strategic challenge for CFOs is consequently not to maximize the level of automation, but to determine where AI creates the greatest economic and organizational value, where human judgment remains essential, and how the two can be combined within an effective finance operating model (Abbas, 2026; Stratopoulos and Wang, 2025).

Ultimately, AI transformation in finance should be evaluated as an organizational capability rather than a technology acquisition programme. Organizations that successfully combine AI with high-quality data, redesigned processes, capable professionals, effective governance, and clear decision rights are more likely to convert technological potential into sustained performance improvements. In this sense, the central competitive advantage is not having AI, but knowing how to organize around AI (Almeida and Santos Júnior, 2025; Abbas, 2026; Stratopoulos and Wang, 2025).

14. Conclusion

Artificial intelligence is changing the economic and organizational conditions under which the finance function operates. The evidence reviewed in this paper suggests that the transformation should not be understood simply as the automation of accounting and finance tasks. Rather, AI is altering the way information is produced, interpreted, governed, and translated into managerial decisions. This distinction is central to understanding both the opportunities and the risks associated with AI adoption.

The analysis of the CFO AI Navigator 2026 alongside recent academic research leads to five principal conclusions.

First, AI transformation in finance is fundamentally a leadership and organizational challenge. The CFO has a distinctive role because AI affects resource allocation, performance measurement, forecasting, internal controls, risk management, process design, and workforce capabilities. While IT and technology functions remain essential to implementation, finance leadership must determine where AI creates economic value, where professional judgment is required, and how risks and benefits should be balanced. CFO ownership should therefore extend beyond sponsorship of individual projects to the design of the broader AI-enabled finance operating model.

Second, AI value is highly context-dependent. The empirical evidence from Noy and Zhang (2023), Brynjolfsson et al. (2025), and Dell'Acqua et al. (2025) demonstrates that AI effects vary across tasks, workers, and contexts. This makes generalized productivity or ROI claims difficult to transfer directly between organizations. The practical implication is that finance leaders should evaluate AI at the level of specific processes and establish organizational baselines before making investment decisions. The relevant question is not whether AI is productive in the abstract, but whether it improves the economics and outcomes of a particular finance process under the organization's actual operating conditions.

Third, AI creates value when technology is combined with process redesign and human expertise. Simply adding AI to an existing process may generate incremental efficiency while leaving underlying structural inefficiencies intact. Greater value may arise when organizations redesign workflows, decision rights, controls, and human responsibilities around new AI capabilities. This supports the paper's central proposition that the appropriate unit of analysis is the AI-enabled finance process rather than the individual technology.

The evidence also challenges a simple substitution narrative. AI can automate repetitive activities, augment analytical work, and potentially support increasingly autonomous processes. Yet the research indicates that AI performance depends on task characteristics and the interaction between technology and human expertise. In high-judgment financial activities, human review, professional skepticism, contextual interpretation, and accountability remain essential. Indeed, as AI makes the production of plausible financial analysis easier, the ability to evaluate and challenge that analysis may become more important.

Fourth, governance is a condition for scaling rather than merely a constraint on adoption. The findings of Almeida and Santos Júnior (2025), Hadley et al. (2025), and Eisikovits et al. (2024) indicate that responsible AI requires organizational structures, leadership support, training, accountability, and integration with existing processes. For finance functions, governance must address data quality and confidentiality, model validation, explainability, security, human accountability, monitoring, and regulatory requirements. These mechanisms become increasingly important as AI moves from advisory applications toward systems capable of influencing or executing material financial activities.

Governance should therefore be embedded from the beginning of the AI lifecycle. Its purpose is not simply to prevent inappropriate use but to create sufficient organizational confidence for successful applications to scale. Without effective governance, technically successful pilots may remain isolated because executives are unwilling to accept their associated operational, regulatory, or reputational risks.

Fifth, AI is likely to transform the composition of finance work rather than simply reduce the need for finance professionals. Abbas (2026) highlights the simultaneous possibilities of automation, upskilling, reskilling, deskilling, and changing professional boundaries. As AI reduces the effort required for information production, finance professionals may increasingly focus on interpretation, scenario analysis, business partnering, strategic decision support, and governance. However, this transition is not automatic. Organizations must deliberately invest in AI literacy, data literacy, professional judgment, process design, and critical evaluation of AI outputs.

These conclusions support the revised five-stage CFO AI transformation model developed in this paper. Diagnosis establishes organizational and process readiness. Prioritization focuses scarce resources on use cases with credible economic and strategic potential. Experimentation tests whether AI produces measurable net value in the organization's actual environment. Process redesign converts successful experiments into redesigned human-AI workflows. Institutionalization embeds AI within the finance operating model through governance, monitoring, portfolio management, workforce development, and continuous reassessment.

Importantly, these stages should not be regarded as a one-time implementation sequence. AI transformation is better understood as a continuous organizational capability cycle. As technologies, processes, risks, and organizational capabilities evolve, finance leaders must repeatedly reassess where AI creates value, where governance requirements have changed, and where human capabilities need to be strengthened.

The paper's research propositions provide a corresponding agenda for future empirical work. In particular, further research should investigate whether CFO ownership improves AI scaling, whether process-based use-case selection predicts realized value, how data and process maturity moderate AI outcomes, when human-AI collaboration outperforms automation, whether governance maturity facilitates scaling, and whether AI shifts finance performance from information production toward decision enablement. Longitudinal studies, field experiments, comparative case studies, and large-sample empirical research will be particularly important for moving the literature beyond demonstrations of technological potential toward evidence concerning sustained organizational outcomes.

Several limitations should nevertheless be acknowledged. The CFO AI Navigator 2026 is a practitioner guide rather than a peer-reviewed empirical study, and its reported ROI figures should consequently be interpreted as practical signals rather than universal benchmarks. Moreover, much of the academic literature on generative and agentic AI remains relatively recent. Evidence concerning long-term organizational effects, professional development, governance effectiveness, and the consequences of increasingly autonomous AI systems is therefore still developing. The propositions advanced in this paper should consequently be regarded as a structured agenda for empirical investigation rather than as established causal relationships.

Despite these limitations, the convergence between the practitioner and academic perspectives is notable. Both point toward a shift away from technology-first adoption and toward disciplined organizational transformation. Both emphasize the importance of measurable value, organizational readiness, human capabilities, governance, and process redesign. Most importantly, both suggest that AI should be understood as a change in the architecture of work and decision-making rather than merely as the introduction of a new class of software.

The central conclusion of this paper can therefore be stated simply: the strategic value of AI in finance does not primarily arise from replacing finance work; it arises from changing how finance work is organized, governed, and connected to managerial decision-making.

For CFOs and boards, this means that AI maturity should not be measured by the number of tools deployed or pilots completed. A more meaningful measure is the organization's ability to identify valuable opportunities, evaluate them rigorously, govern them responsibly, redesign processes around human-AI collaboration, and continuously develop the capabilities required to use AI effectively.

The future finance function is therefore unlikely to be either fully automated or unchanged. It will increasingly be a hybrid, AI-enabled and human-governed system in which machines perform more information-intensive and repetitive activities while finance professionals concentrate relatively more on judgment, interpretation, challenge, communication, governance, and strategic decision support.

Ultimately, the competitive question is not which organization adopts the most AI. It is which organization learns most effectively how to combine AI capabilities with financial expertise, high-quality data, mature processes, effective governance, and human judgment. For CFOs, that is the transition from AI experimentation to genuine finance transformation.

References

Abbas, K. (2026). Management accounting and artificial intelligence: A comprehensive literature review and recommendations for future research. The British Accounting Review, 58(2), 101551.

Almeida, P. G. R. de, & Santos Júnior, C. D. dos. (2025). Artificial intelligence governance: Understanding how public organizations implement it. Government Information Quarterly, 42, 102003.

Dong, M. M., Stratopoulos, T. C., & Wang, V. X. (2024). A scoping review of ChatGPT research in accounting and finance. International Journal of Accounting Information Systems, 55, 100715.

Eisikovits, N., Johnson, W. C., & Markelevich, A. J. (2024). Should accountants be afraid of AI? Risks and opportunities of incorporating artificial intelligence into accounting and auditing. Accounting Horizons, 39(2), 1–7.

Hadley, E., Blatecky, A., & Comfort, M. (2025). Investigating algorithm review boards for organizational responsible artificial intelligence governance. AI and Ethics, 5, 2485–2495.

Noy, S., & Zhang, W. (2023). Experimental evidence on the productivity effects of generative artificial intelligence. Science, 381(6654), 187–192.

Stratopoulos, T. C., & Wang, V. X. (2025). Artificial intelligence and accounting research: A framework and agenda. International Journal of Accounting Information Systems, 56, 100760.

Brynjolfsson, E., Li, D., & Raymond, L. R. (2025). Generative AI at work. The Quarterly Journal of Economics.

Dell'Acqua, F., et al. (2025). Navigating the jagged technological frontier: Field experimental evidence of the effects of artificial intelligence on knowledge worker productivity and quality. Organization Science.

Parnova Consulting. (2026). CFO AI Navigator 2026: Practical guide for CFOs, finance leaders and boards. Practitioner guide.

Contact

Reach out via email for inquiries.

Email

Subscribe to newsletter

info@grcadvisory.ch

© 2025. All rights reserved.