Research Paper
Why AI Capability Is Outpacing Leadership, Trust and Organisational Capacities
Most organisations are not failing at AI because of technological issues.
Many organisations now have more capability than they can meaningfully use. Foundation models that approximate years of specialist expertise are available on demand. Cloud infrastructure has collapsed the cost of deployment. Data is plentiful. Talent is available for hire. The technical barriers that once defined the AI conversation have largely fallen.
And still, value creation falls short.
McKinsey reports ~80% of companies adopting AI see no significant impact on financial performance.²
BCG's 2025 study of 1,250 enterprises found only 5% generate substantial value at scale, while 60% extract minimal returns despite continued investment.⁴
RAND finds more than 80% of AI projects fail to reach meaningful production deployment.38
The constraint is not capability. It is alignment.
When AI initiatives underperform, it is rarely because the model is wrong. It is because the organisation never agreed on what the system is for, who it is supposed to serve, how its outputs would be acted upon, or what value it would actually bring once it is running. This paper calls that condition the AI Value Alignment Gap: the distance between what AI is expected to deliver and what it produces when such questions go unresolved.
[box-info] This paper sets out a four-layer framework spanning Technology Capability, Leadership Alignment, Organisational Capability, and Societal Trust. The argument is direct: before organisations invest further in AI capability, they must invest in alignment. Without it, additional capability will deepen the gap rather than close it. [box-end]
We are passing through one of the most consequential technological transitions in modern history. Breakthroughs in large-scale models, multimodal systems, and agentic architectures have expanded AI's technical capability. The Model Evaluation and Threat Research project finds that AI is becoming roughly twice as capable every seven months.¹ Tasks that required years of specialist expertise can now be approximated in seconds. Industries from healthcare to finance to manufacturing are being remade.
AI now dominates board agendas. Technology budgets have been redirected toward AI infrastructure. Functions including strategy, operations, customer experience, and risk are being rebuilt with AI at their centre. [box-end]
For every organisation that has embedded AI and seen genuine returns, many more have launched initiatives, committed substantial resources, and produced limited sustained impact. The empirical record is consistent and troubling. A large cross-country executive survey shows that around 80% of firms using AI report no measurable productivity gains.³ Only 5% of firms have achieved transformative bottom-line value from AI at scale.⁴ Studies indicate that only a small fraction of AI initiatives move beyond pilot stages,⁵ with many remaining trapped in proofs-of-concept.⁶,⁷
At the macro level, substantial task-level gains fail to aggregate into measurable economic impact. The value gap, between the top and the poorest-performing firms, widens further as leading firms reinvest AI-generated returns into improved capability, producing compounding competitive advantages.⁴ [box-end]
Surveys of senior executives reveal a consistent pattern: high confidence in AI strategy, low confidence in AI delivery. This reflects something fundamental: a widespread failure to align the actors, priorities, and expectations that determine what AI value looks like for the organisation. [box-end]
The organisations succeeding with AI are not those with the most advanced models. They are those that have achieved alignment between technology, leadership, people, and purpose. Without such alignment, continued investment risks reinforcing a cycle of technological progress without corresponding economic or societal benefit. [box-end]
This paper diagnoses that failure and offers a more rigorous foundation for closing the gap between AI ambition and AI impact. It is addressed to senior executives, board members, policymakers, and practitioners who sense that something important is missing from current conversations about AI strategy.
Consider what organisations now have access to. Foundation models from OpenAI, Anthropic, and Google DeepMind can reason, generate professional-quality text, analyse images, write code, and make predictions across domains once exclusive to highly trained specialists. Cloud platforms deliver these capabilities on demand at a fraction of the cost of traditional development. Data has never been more plentiful. Most organisations of any scale generate more operational and customer data than they could have analysed a decade ago.⁸ Storage is cheap. Compute capacity has expanded dramatically. Tooling has matured. Yet across sectors, the translation of these inputs into measurable economic or strategic value remains highly uneven.
This tension between technological abundance and limited impact realisation constitutes the AI Capability Paradox. The conditions necessary for AI success (data, compute, models, talent, capital) are increasingly available and affordable. And yet organisations consistently report that translating these inputs into sustained, meaningful value remains elusive.
Only a small minority of organisations deploying AI at scale captured significant financial value.⁹ [box-end]
Just 1% of companies consider themselves "mature" in AI deployment, with 30-40% of potential impact eroded by fragmented systems and weak incentives.¹² Only about 6% of firms achieve meaningful financial returns at the EBIT level.¹³ [box-end]
Documented the persistent gap between AI pilots and AI in production.¹⁰ [box-end]
Many organisations investing heavily in AI capability reported flat or negative returns on that investment.¹¹ [box-end]
Organisations are accumulating capability stock - data, models, and tools - without developing the organisational capacity required to deploy them effectively. The consequence is a widening gap between what AI systems can theoretically deliver and what institutions are practically able to extract.
That gap reflects failures not in the technology stack, but in the human and institutional structures around it: how leaders define success, how organisations build capacity for adoption, and how trust is established with the people most affected. Understanding this gap requires a clearer analysis of how different actors define "value" in the first place, and why those definitions rarely intersect.
The word "value" is used constantly in discussions about AI. Yet it means very different things depending on who is speaking, and those differences matter enormously for how AI initiatives are designed, funded, governed, and judged. Fragmentation across stakeholders is a persistent barrier to AI value realisation.
[box-info] Organisations frequently struggle to scale AI not because models fail, but because business units, leadership teams, and technical functions are not aligned on what success looks like. [box-end]
These are not simply different priorities that can be reconciled through better communication. They reflect genuinely different theories of what AI is for and who it should serve.
Boards take a distinct position. Unlike executives, they are not responsible for delivery, but for oversight.¹⁴ They ensure AI investments align with fiduciary duties and protect organisations from systemic risk.¹⁴
From a board perspective, AI value is inseparable from accountability. Systems must perform, and do so in ways that are auditable, explainable, and aligned with long-term strategy. This introduces a moderating force into AI adoption. While CEOs may prioritise speed and competitive differentiation, and CFOs focus on near-term returns, boards evaluate value across longer time horizons under stricter risk constraints.15
Evidence of this shift is visible. EY Center for Board Matters research found that nearly half of Fortune 100 companies now cite AI risk as part of board oversight responsibilities, a threefold increase from 16% in 2024 to 48% in 2025.16 Around 40% of companies now disclose that at least one board-level committee has been charged with AI oversight, almost four times compared to the 11% in 2024.17 [box-end]
CEOs typically frame AI in terms of strategic positioning - competitive advantage, market differentiation, long-term repositioning. Kearney's 2025 study finds that 62% of CEOs view AI as a powerful shift that will reshape business.18 A Gartner survey of 456 CEOs and senior business executives found 77% believe AI is opening a new business era.19
From this perspective, AI is valuable insofar as it reshapes industry structure or unlocks new growth. Yet only 8% of companies prioritise positioning themselves for value creation beyond the next three-year horizon, suggesting that even among strategically oriented leaders, long-term thinking about AI value remains rare. 20 [box-end]
Finance leaders interpret value through quantifiable return - cost reduction, productivity gains, capital efficiency. L.E.K. Consulting's 2025 Office of the CFO survey shows CFOs review AI gains in terms of productivity, work quality, and cost reduction.21 The World Economic Forum's CFO guidance focuses on cost and productivity.22 Strategic AI, in this view, has to be reflected in the profit and loss statement, improving operating income and earnings per share.23 [box-end]
Beyond leadership circles, fragmentation deepens. The dominant employee sentiment is anxiety. Pew Research finds 52% of US workers are worried about AI's future workplace impact, and 32% believe it will reduce their long-run job opportunities.25 The anxiety intensifies with exposure: employees at organisations undergoing comprehensive AI-driven workflow redesign are more worried about job security (46%) than those at less-advanced companies (34%).26
On fairness, employees are conditionally open. A Resume Now survey found 66% believed AI-led management would make workplaces fairer, but 85% said greater AI transparency would be required to increase trust.27 McKinsey's 2024 employee survey found cybersecurity risks (51%), inaccuracies (50%), and personal privacy (43%) as top concerns about generative AI at work.28
Where employees do find value, it is tied to tangible improvements. PwC's 2025 Global Workforce Hopes & Fears Survey of nearly 50,000 workers found that daily generative AI users are far more likely than infrequent users to report gains in productivity (92% vs 58%) and job security (58% vs 36%).29 Yet access remains unequal: only 14% of the global workforce uses generative AI daily, and just 51% of non-managers feel they have adequate learning resources, compared to 72% of senior executives.30
Employees define AI value not by what the technology can do, but by what it does for them. [box-end]
Customers evaluate AI based on trust, transparency, and experience quality. In the UK, 72% of clients demand ethical AI practices from financial advisors using AI tools.31 A KPMG 2024 Generative AI Consumer Trust Survey found that over 80% of consumers expect companies to conduct regular internal audits of AI systems for bias, fairness, and security, and to establish clear ethical frameworks for AI use.32 [box-end]
Regulators operate where policy intent meets operational reality. Authorities such as the Federal Trade Commission,33 the Information Commissioner's Office, and the European Data Protection Board34 evaluate AI systems through compliance, oversight, and risk mitigation. For regulators, AI value is defined by whether systems can operate safely, lawfully, and accountably, adherence to standards around fairness, transparency, explainability, and data protection. Value, in this context, is inseparable from the ability to withstand investigation, audit, and potential enforcement action. [box-end]
Policymakers operate at a different level. Where regulators enforce, policymakers define direction. They provide a normative compass for developers, deployers, and users of AI.35
For them, AI value is not primarily organisational or technical, but societal and economic. Institutions such as the European Commission and legislative bodies frame AI in terms of national competitiveness, economic growth, labour market impact, and public welfare. An AI system that is profitable and technically robust may still be constrained if it conflicts with labour protections, competition policy, or democratic norms. [box-end]
Communities - civil society organisations, social movements, populations affected by AI - engage in civic participation as a way of asserting their understanding of AI value.36 Organisations such as the Electronic Frontier Foundation and multi-stakeholder initiatives like the Partnership on AI illustrate how societal actors shape expectations.
For communities, value is defined less by performance or profit and more by lived experience: fairness, inclusion, dignity, and agency. Whether an AI system creates value depends on how it affects real people in practice, not how it performs in controlled conditions. [box-end]
[box-info] Individually, each perspective is rational and legitimate. Collectively, they create systemic misalignment. Governance structures designed around technical metrics fail to capture trust considerations. Projects optimised for short-term financial returns miss the long-term organisational investment required for adoption. Employees disengage when AI feels like an imposition rather than an enhancement. Regulators intervene when safety and fairness are treated as afterthoughts.
This fragmentation is a root cause of the AI Value Alignment Gap. Organisations invest heavily in capability while the actors involved hold fundamentally different expectations about what that capability should produce, and those expectations are rarely surfaced, examined, or resolved before deployment begins.
The AI Value Alignment Gap is not a technology failure. It is a failure of shared understanding about what AI is for, and whose definition of success should prevail. [box-end]
Closing the gap requires more than improved project management or better data pipelines. It requires a structured approach to aligning four interdependent layers of any AI initiative. AI value is fully realised only when all four are coherent and reinforcing.
The AI Value Alignment Framework is deliberately simple. Its purpose is not to map every dimension of AI complexity, but to provide a practical diagnostic, a tool leadership teams can apply when assessing why an AI initiative is underperforming, or before launching one.
The layers are nested by design. Technology Capability sits at the centre because it is the most bounded, operating within the space the surrounding layers permit. Societal Trust forms the outermost boundary because it sets the conditions under which everything else is allowed to function. Value accumulates progressively as each layer comes into alignment with the others. When all four are coherent and reinforcing, AI initiatives generate outcomes that are technically reliable, strategically directed, operationally embedded, and publicly legitimate.
[box-warning] BCG's 2025 study of 1,250 enterprises found 60% of organisations generate no material value from AI despite continued investment, and only 5% create substantial value at scale.37 RAND's 2024 analysis found more than 80% of AI projects fail to reach meaningful production deployment.38 [box-end]
Technology enables. Leadership defines. Organisation translates. Society legitimises. Alignment determines whether capability becomes value.
Technology capability is the core layer that is most visibly associated with AI projects and typically the first area of investment. It encompasses the quality of models, the robustness of data infrastructure, and the ability to iterate. The progress here is real.
Stanford HAI AI Index 2025 documents dramatic improvements on MMMU, GPQA, and SWE-bench benchmarks.39 [box-end]
Inference cost for GPT-3.5-level systems fell more than 280-fold between November 2022 and October 2024.40 [box-end]
Hardware costs have declined roughly 30% annually, while energy efficiency has improved 40% each year. [box-end]
By 2025-2026, agents with reasoning capabilities can plan, call tools, and complete complex tasks without constant human management.41 [box-end]
But technology capability alone creates possibility rather than value. There is a persistent and costly gap between what AI systems can do during testing and what they deliver in practice. Pilots succeed in controlled environments. In production, reality can intervene in the form of messy real-time data, legacy system integration, and decision-making challenges, and infrastructure that, although it worked in the demonstration, cannot support the system at scale.42
This pattern is well understood by those leading transformations in complex environments. As David Edem, Data & AI Transformation Executive, observes from his work across energy and industrial software:
David Edem - Data & AI Transformation Executive
In my experience, leading AI and data transformation across energy and industrial software, the limiting factor is rarely the technology. It is organisational will: the readiness of leadership to treat AI as a business transformation lever, not a technology programme.
The implication is that capability is necessary but never sufficient. The other three layers do the work of converting capability into value.
Leadership alignment is the most overlooked layer. When executive teams do not share a coherent view of what AI should achieve and how success will be measured, AI initiatives become politically contested, chronically under-resourced, or perpetually deferred. Achieving alignment requires explicit, structured conversations about competing definitions of value before significant investment begins. Without it, the organisation cannot make consistent decisions about what to build, what to defer, and what to govern.
BCG found nearly 100% of "future-built" organisations report deeply engaged C-suites, vs only 8% in lagging companies.
Future-built companies achieve 1.7x more revenue growth than slower competitors.
3.6x greater three-year total shareholder returns for aligned organisations vs laggards.
[box-info] McKinsey's 2025 report identifies leadership misalignment as among the most challenging operational headwinds facing AI adoption, and concludes that the biggest barrier to success is leadership, not employees.43 [box-end]
Leadership alignment is not a soft consideration. It is a financial performance variable.
The Founding Council voiced this point with particular force. Where leadership accountability for AI outcomes is held by a senior business owner working in active partnership with a Chief Data and AI Officer, trade-offs get made and execution follows. Where AI is seen as a technology function alone, even well-designed initiatives stall at proof of concept.
David Edem - Data & AI Transformation Executive
That partnership only works when data is actively managed as a product, with clear ownership, quality, and accessibility aligned to business priorities. Even then, the deeper constraint is rarely the data itself. It is whether leadership is willing to act when the answer is inconvenient. In that sense, the value alignment gap is, in practice, a leadership alignment gap.
This reframing - that the value alignment gap often reflects a leadership alignment gap - can help get to the source of why technical investment alone consistently underperforms.
Organisational capability is where AI strategy most frequently collides with operational reality. An organisation may have excellent models and aligned leadership, but if the people expected to use those models lack the skills, the processes, or the cultural permission to act on AI outputs, value will not materialise.
IDC estimates AI skills shortages may cost the global economy up to $5.5 trillion by 2026.44
The World Economic Forum reports 94% of leaders face shortages in AI-critical roles.45
Only 51% of frontline employees are regular AI users, a figure that has stagnated.26
Just 46% of organisations integrate workforce planning into their AI roadmaps.45
The Council framed the same point through the lens of execution discipline. Amy Shi-Nash, Chief Analytics & Data Officer, argues that failure or success of AI initiatives are related less to the technology itself and more to how the problem is defined and how the system is embedded.
Amy Shi-Nash - Chief Analytics & Data Officer, Professor Monash University
High-impact projects start with clear, business-relevant use cases, are built on reliable data, and are designed with end users in mind, ensuring people actually use them. When initiatives fall short, it's often due to vague objectives, poor integration into day-to-day operations, or lack of user trust and adoption, leaving technically sound solutions underutilised.
Execution discipline depends on a small number of foundational conditions. Ming Tang, Chief Data and Analytics Officer, identifies three that consistently determine whether organisations move from capability to value: a clear business-relevant use case that produces visible early results; data foundations that have been actively prepared, with access secured, governance completed, and the location of value within the data understood; and the readiness of a multidisciplinary team to redesign processes as AI is integrated, rather than layering AI onto legacy operations.
Ming Tang - Chief Data and Analytics Officer, Former NHS England
Having a clear use case that is structured to solve a real business problem and therefore deliver value quickly. The data teams have understood and prepared the ground… Preparation for re-imagining the potential future process with AI integration.
Embedded in this is a more specific operational concern. As Lambert Hogenhout, who leads on Data, AI, Privacy and Responsible Tech at the United Nations, observes:
Lambert Hogenhout - United Nations
The main determining factor in success of AI initiatives I see in our organisation is the extent to which there is clarity and agreement about the expected output and how that will contribute to larger processes or workflows.
Clarity about expected output is not the same as having a use case, nor a matter of technical specification. It is an organisational condition: a shared understanding across the people who commission, build, deploy, and use the system. Where that agreement is absent, the output may be delivered but not acted upon.
The design dimension carries equal weight. Julie Wall, Professor at the University of West London, emphasises that AI value resides in how capability is connected to human judgement within a defined context, with explicit links between inputs, outputs, and accountable use:
Julie Wall - Professor at the University of West London
AI initiatives tend to be problematic when value is perceived to come from the technology itself, rather than from how it is integrated and used in context. The gap is often not technical, but organisational, when there is a failure to connect model outputs to accountable use, including who is responsible for acting on and validating those outputs.
Even where execution and design are sound, organisations frequently lack the mechanisms to detect, in real time, how AI systems are actually being experienced by users. Constantinos K. Coursaris, Professor at HEC Montréal, makes the case for rigorous, mixed-method feedback loops that combine behavioural, self-reported, and increasingly physiological signals to diagnose where value is created or eroded. Deploying AI faster than an organisation can learn from its deployment is, in his framing, a form of failure:
Constantinos K. Coursaris - Professor at HEC Montréal
The 'value alignment gap' is less about the technology itself, and more about the organisation's ability to remain deeply anchored in human experience while scaling intelligence.
Governance is the connective tissue across all of this. Gartner research shows that 45% of organisations with high AI maturity keep their AI initiatives operational for at least three years, compared with only 20% among lower-maturity peers - with dedicated governance structures, leadership accountability, and lifecycle oversight as primary differentiators.47
Peggy Tsai, Executive Director at JPMorganChase, calls for an institutional perspective, where AI initiatives have to achieve measurable outcomes for the business, and governance structures deliver both guardrails and value definitions to enhance organisational agility to upskill, reskill, and operate cross-functionally:
Peggy Tsai - Executive Director at JPMorganChase
AI initiatives must have measurable impact, solving a specific and measurable business problem. Another foundational aspect is the creation of an AI Governance Council to set guardrails and define value across the company.
Without these, even strong capability investments produce inconsistent, unsustainable results.
[box-info] NTT DATA's analysis identifies the core adoption problem clearly: if employees do not trust a concept or tool, not only will they fail to use it, they will actively work against it.46 Trust is is required at various levels: employees need to trust their organisation, the people governing the AI models, and the outputs of the models themselves. [box-end]
Societal trust is one of the most underestimated source of AI risk and the most difficult to recover once lost. It can manifest at multiple levels at once: employees need trust to counteract fears that AI may diminish their roles; customers need confidence that systems are fair and safe; and regulators and civil society must be satisfied that deployment is responsible. Organisations that treat trust as a communications challenge rather than a design and governance priority consistently find it becoming a limit to what they can achieve.
Trust AI in China (Edelman 2025 Trust Barometer)48 [box-end]
Trust AI in the United States (Edelman 2025 Trust Barometer)48 [box-end]
Trust gap between employees who found AI helpful vs those who did not (Edelman Flash Poll)⁵⁰ [box-end]
The internal trust dimension is equally consequential. An Aberdeen study found that 70% of Baby Boomers, 63% of Generation X, and 57% of Millennials and Generation Z agree AI will put jobs at risk.49 The resulting anxiety has produced active resistance to AI integration, especially when organisations fail to address it through transparent communication, reskilling, and credible governance.
For Council members, the trust question is reflective of a deeper issue, the breakdown in cross-organisational communication. Lindley Gooden, author of The Future of Truth (and How to Get There), argues that AI is far more than a technical challenge:
Lindley Gooden - Author, The Future of Truth (and How to Get There)
If we're going to make AI succeed, better communication and understanding is at the heart of it. When the flow of information breaks down, the process breaks down. We need something quite scary: honesty. Without clarity, AI projects fail. Existing communication cracks are transplanted onto tech solutions, and customers spot them.
It is a matter of structural alignment. Most organisations already carry fractures between leadership and staff, between functions, between the business and its customers. When AI is deployed into those fractured environments, the technology does not heal the breaks. It inherits them and scales them.
[box-check] Trust is not fixed. It is built through demonstrated benefit, transparent governance, and meaningful inclusion in how AI is designed and deployed. Organisations that treat societal trust as an operational imperative are those best positioned to sustain AI value over time. [box-end]
The AI Value Alignment Gap is not an abstract concept. It manifests in recognisable patterns across organisations. These are commonly misdiagnosed as technical limitations or data issues, but they are actually symptoms of misalignment across the four layers.
Pilots succeed in controlled conditions but frequently fail to scale. Scaling requires organisational integration, leadership alignment, operational readiness, and stakeholder trust, none of which the pilot stage is designed to test. Success of a pilot is not evidence of organisational readiness. It is often where failure begins. [box-end]
In many organisations, AI activity is highly visible (announcements, prototypes, demonstrations, governance frameworks) without measurable impact. This pattern is driven by competitive pressure, the absence of clear value definition, and fragmented leadership priorities. Activity is mistaken for value; visibility is mistaken for impact. [box-end]
Where value is not defined consistently across the organisation, AI amplifies existing fragmentation rather than resolving it. Different functions optimise for different outcomes. Success metrics conflict. AI exposes misalignment that already exists; it does not fix it. [box-end]
Even when systems perform well, they may not be adopted because of weak transparency, unclear accountability, perceived risk, or regulatory uncertainty. Lack of trust then leads to ceiling on scale. [box-end]
AI requires coordination across technology, business units, risk and compliance, and leadership. Where this breaks down, organisations produce duplicated efforts, inconsistent standards, slow decision-making, and stalled deployment. That's when AI is a coordination problem rather than a single-function problem. [box-end]
Where the necessity of AI deployment has not been clearly defined or lacks direction at the outset, organisations may end up managing high-cost experiments in anticipation of forward momentum, but without a clear sense of purpose. [box-end]
Markus Krebsz, EW-AiRM creator, UN ECE WP.6 AI Project Lead, and Honorary Professor at Stirling University, has been asking board members, senior executives, and risk leaders the same question for years:
Do you actually know why your firm is deploying AI? The responses are rarely reassuring and often unsurprisingly vague. A large majority of organisations are pursuing AI because it feels strategically necessary, and they are hoping the value will become more apparent once the systems are live. This is the value alignment gap in its most simple form, and one of the most reliable predictors of whether an AI initiative will deliver meaningful value or quietly disappear as another expensive, yet avoidable lesson learnt.
Where organisations are not aligned on the intended purpose of AI systems, the speed, scope, and manner of deployment can outpace the governance structures needed to oversee them responsibly. In these conditions, ethical and compliance failures emerge as structural consequences of misalignment.
Systems deployed without clear accountability, shared definitions of acceptable use, or operational readiness are more likely to generate biased outcomes, inconsistent decision-making, opaque accountability chains, and regulatory exposure.
As legal and governance frameworks surrounding AI remain comparatively immature, organisations that treat ethics and compliance as downstream considerations rather than integrated deployment conditions risk creating systems that are technically functional but institutionally unstable. [box-end]
These patterns are not separate issues. They are interconnected symptoms of the same underlying problem: misalignment across the value system. Addressing them individually treats symptoms, not causes, which is why organisations frequently fix one issue only to encounter another.
The AI Values Institute convened members of its Founding Council (senior executives, practitioners, and researchers who have led AI initiatives across industries, sectors, and institutions) and posed a single question:
What follows is a synthesis of their responses, organised around seven recurring themes. Together they form a cross-sector convergence: independent observations from very different vantage points that essentially arrive at the same diagnosis.
Across the Council, the most consistent message was that AI failures rarely originate in the technology stack. They originate in the quality of leadership.
David Edem describes the pattern: where a senior business leader owns organisational outcomes in active partnership with the Chief Data and AI Officer, execution follows; where AI is delegated to the technology function, even well-designed initiatives stall.
Markus Krebsz frames the consequence: governance written but never enforced, pilots that never scale, and narratives built for external optics rather than operational reality.
Each of these voices reinforces the same conclusion. The AI Value Alignment Gap is, at its core, a leadership alignment gap. [box-end]
Organisations consistently overestimate how prepared they are to deploy AI meaningfully.
Amy Shi-Nash observes that high-impact initiatives begin with clear, business-relevant use cases, reliable data, and built-in considerations of end users.
Ming Tang formalises this into three preconditions: a structured use case, prepared data foundations, and a multidisciplinary team capable of redesigning processes around AI rather than layering AI onto legacy ones.
Markus Krebsz adds a sharper note of caution: readiness is not a function of budget or headcount but of AI literacy, governance maturity, data discipline, and cultural conditions. The digital skills gap means many organisations are deploying AI systems not fully understood by their own risk, legal, and compliance functions. [box-end]
A third theme concerns the absence of a clear, shared definition of value at the point of deployment. Markus Krebsz argues that the first question every organisation should ask is whether AI is genuinely the right tool for the problem - what he terms a necessity assessment - before any technology decision is made. Without it, organisations end up managing not much more than high-cost experiments.
Peggy Tsai reinforces the same point from inside a major financial institution: AI initiatives must have measurable impact tied to specific, measurable business problems. Ming Tang places use case clarity as the first precondition for value, ahead of any technical work.
These voices reinforce that value has to be treated not as something AI produces but as something the organisation must define, agree on, and govern in advance.
The fourth theme is that AI value depends on how well organisations communicate, internally with employees and externally with customers, and on how honestly they engage with what AI is and is not delivering.
Lindley Gooden makes the case that communication is the hidden infrastructure of AI: existing communication cracks within the organisation are transplanted onto AI deployments and customers spot them. Constantinos Coursaris extends the same logic to user experience: where organisations deploy faster than they can learn, technical sophistication outpaces understanding, and user trust collapses. Amy Shi-Nash links this directly to adoption: technically sound solutions remain underused when user trust is absent.
Trust is not a downstream outcome of AI deployment success. It is a prerequisite for it.
Peggy Tsai argues for governance councils that set guardrails and define value across the enterprise.
Markus Krebsz frames governance as the most consistent differentiator - accountability structures that are explicit, documented, and actually enforced; multidisciplinary teams including risk, legal, ethics, and customer expertise; and an ethics filter applied at every decision juncture.
Bob Enofe brings these threads together at the strategic level, arguing that successful AI investments embed AI into the organisation's purpose, culture, and accountability structures from the outset rather than treating value alignment as a one-time implementation issue. [box-end]
[box-light] 6. Human Judgement As the Unit of Value
AI value resides in how capability connects to human decision-making.
Julie Wall makes the case for structural support: AI initiatives succeed when designed to support and extend human judgement within a clearly defined context.
Lambert Hogenhout names the operational manifestation: the determining factors are clarity and agreement about expected output and how it contributes to larger processes.
Constantinos Coursaris links this to experience: when AI is used not to automate but to augment understanding and accelerate the path from insight to action, initiatives are far more likely to deliver meaningful value. [box-end]
Bob Enofe - Board of Directors, Innovators & Entrepreneurs Foundation
The gap between AI's potential and its realised value is rarely technical. It is fundamentally informed by human failure, and shaped by whether the leaders of initiatives have both the clarity to define what meaningful value entails for stakeholders and the institutional courage to reorganise incentives, workflows, and decision rights around the definition.
Since AI initiatives cross departmental boundaries, they need to be structured accordingly. David Edem advocates for partnerships between senior business leaders and technical teams. Bob Enofe focuses on reorganising incentives, workflows, and decision rights; Lindley Gooden observes that AI deployments inherit and amplify communication fractures. Each perspective points to the same structural reality: AI success depends as much on coordination as it does on technology. [box-end]
[box-check] These seven themes reflect cross-sector convergence among practitioners working in financial services, energy, technology, the public sector, the United Nations, academia, and governance. The fact that they arrive at the same diagnosis from different vantage points is evidence the AI Value Alignment Gap is a structural feature of how organisations are currently approaching AI. [box-end]
The AI Value Alignment Gap is, ultimately, a leadership challenge. It cannot be resolved by technology teams in isolation, by consultants producing strategies that go unread, or by governance frameworks built after deployment has already begun. Closing the gap requires sustained, informed leadership engagement across all four layers of the alignment framework. Four imperatives follow.
Before an organisation commits further resources to AI capability, leadership must answer a prior question: what does value mean, in this context, for these stakeholders? This is an ongoing, structured conversation that surfaces competing definitions, establishes shared measures of success, and creates accountability for delivering them. Practically, it means involving the full range of value stakeholders in the definition of success criteria from the outset, making explicit trade-offs between short-term financial return and longer-term trust-building, and building measurement frameworks that reflect the full complexity of value, not only the dimensions that are easiest to quantify. [box-end]
Leadership misalignment is one of the most consistent predictors of AI initiative failure. When executive teams disagree about what AI should achieve, that disagreement cascades through every layer of the organisation. Resource allocation becomes contested. Governance frameworks become politicised. In these conditions, ethical and compliance risks often emerge long before organisations have developed the institutional capacity to manage them. Achieving alignment requires more than an agreed strategy document. It requires ongoing governance structures that bring executive perspectives into regular structured dialogue, clear accountability for AI outcomes at the board level, and explicit mechanisms for resolving the trade-offs AI deployment inevitably surfaces. [box-end]
What most commonly triggers failures in AI deployment is the gap between what AI systems can do and what organisations can absorb, apply, and improve. Building organisational capability requires sustained investment in people through ongoing programmes of skill development, process redesign, and cultural change. It requires governance frameworks that make AI outputs legible to those expected to act on them, and that preserve human judgement where it matters most. And it requires leaders who model the behaviours they want to see: engaging with AI tools themselves, asking hard questions about how decisions are being made, and creating the psychological safety for people to raise concerns about AI performance or ethics without fear of reprisal. [box-end]
One of the most persistent sources of the AI Value Alignment Gap is a measurement deficit. Organisations measure what is easy to count (e.g., model accuracy, processing speed, cost per transaction) and systematically neglect the dimensions of value that are harder to quantify but often more consequential: employee trust in AI systems, customer confidence in AI-enabled decisions, the quality of AI governance processes, and the organisation's capacity to detect and correct AI errors over time. Closing this gap means investing in value frameworks that span all four layers, tracking not just whether AI systems perform well but whether they are trusted, adopted, governed, and continuously improved. [box-end]
The AI Values Institute was founded on the recognition that the challenges described in this paper will not be resolved by any single organisation, sector, or discipline working in isolation. They require sustained, collaborative effort across institutions, industries, and geographies. The Institute exists to provide the intellectual infrastructure for that effort. Its work is organised around three priorities.
There is no shortage of AI research. Most of it focuses on technical performance rather than on the human and institutional conditions that determine whether AI actually creates value. The Institute is committed to building a rigorous, empirically grounded understanding of the latter: how value alignment is achieved in practice, what governance structures support it, how leadership behaviour shapes AI outcomes, and how different stakeholder perspectives can be integrated into coherent frameworks for AI deployment.
This research draws on case studies, executive interviews, and cross-sector analysis. It is designed for accessibility as well as rigour, intended for senior executives, board members, and policymakers who make consequential decisions about AI in real time. [box-end]
The AI Value Alignment Gap is, at its core, a leadership accountability problem. Leaders who cannot articulate a coherent definition of AI value, who have not achieved alignment across their executive teams, and who have not invested in the organisational capabilities necessary for AI adoption are failing in their organisation's AI journey.
The Institute develops standards, frameworks, and leadership development programmes that raise the quality of AI governance at board and executive level, including practical tools for assessing leadership alignment, guidance for boards on AI oversight, and programmes that equip senior leaders to engage with AI with the same depth and rigour they bring to financial, operational, and reputational risk. [box-end]
The AI Value Alignment Gap reflects broader tensions between the pace of technological change and the capacity of institutions (e.g., regulatory bodies, civic organisations, democratic processes) to shape that change in ways that serve the public interest. The Institute operates at the intersection of these domains: convening conversations across sectors, contributing to policy debates, and building the connections between technologists, business leaders, civil society, and governments that are necessary for AI governance to keep pace with AI capability. [box-end]
[box-check] This is, ultimately, the purpose that animates the AI Values Institute: not to slow the development of AI, but to ensure that its development serves the values that organisations and societies actually hold. [box-end]
AI capability will continue to advance. Models will improve. Infrastructure will scale. Access to tools and talent will expand. None of this, in itself, will close the gap between AI capability and AI value. In many organisations, it will widen it.
The organisations that succeed in AI will not be those with the most advanced technology. They will be those that achieve alignment - alignment between leadership intent and organisational execution, between technical capability and defined outcomes, between organisational priorities and societal expectations. That alignment will determine which systems are built, how they are deployed, who benefits from them, and whether their outcomes are trusted and sustained.
references
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Suspendisse varius enim in eros elementum tristique. Duis cursus, mi quis viverra ornare, eros dolor interdum nulla, ut commodo diam libero vitae erat. Aenean faucibus nibh et justo cursus id rutrum lorem imperdiet.
