AI Spending Is Not Strategy: Why Executives Must Govern AI Before AI Governs the Organisation

Artificial intelligence is entering the enterprise at a scale that would normally imply strategic clarity. Capital is flowing. Boards are asking for acceleration. Vendors promise transformation. Executives are under pressure to demonstrate that their organisations are not falling behind. Yet the empirical record is now difficult to ignore: much enterprise AI activity is not translating into measurable business value.

The problem is not simply technological immaturity. It is an executive decision-making problem.

The danger is not that organisations are experimenting with AI. Experimentation is necessary. The danger is that experimentation is being mistaken for transformation, expenditure for strategy, and automation for accountable judgement. That is strategy theatre: visible executive activity — pilots, announcements, partnerships, dashboards, transformation language, and capital allocation — that creates the appearance of strategic control without the decision discipline required to govern operational consequence.

The data justifies a more sober managerial question. Not “How fast are we adopting AI?” but: what decisions are we authorising, what consequences are we creating, and who remains accountable when the system acts?

MIT NANDA’s 2025 GenAI Divide report found that despite $30–40 billion in enterprise generative AI investment, 95% of organisations were getting zero return, with only 5% of integrated pilots extracting significant value. Gartner reported that in a May 2025 survey of 506 CIOs and technology leaders, 72% said their organisations were breaking even or losing money on AI investments. S&P Global Market Intelligence found that the share of companies abandoning most AI initiatives before production rose from 17% to 42% year over year, with organisations scrapping an average of 46% of proof-of-concept projects before broad adoption. IBM’s 2025 CEO study reported that only 25% of AI initiatives had delivered expected return on investment, and only 16% had scaled enterprise-wide. McKinsey’s 2025 global AI survey found that while AI use is widespread, just 39% of respondents reported EBIT impact at the enterprise level.

These numbers do not prove that AI lacks value. They prove something more important for executives: AI value is not created by adoption alone. It is created by disciplined integration into business models, workflows, data systems, accountability structures, and decision rights.

Three pathologies of AI strategy theatre

The current enterprise AI problem can be understood through three connected pathologies.

The first is strategy theatre itself: the visible performance of transformation without the decision discipline needed to control consequence.

The second is pilot purgatory: the condition in which organisations authorise experiments more easily than they authorise accountable deployment. Pilots proliferate, demonstrations impress, productivity anecdotes accumulate, and yet few initiatives move into governed operating capability. S&P Global’s finding that 42% of companies abandoned most AI initiatives before production is not only a technology-adoption statistic. It is a decision-system signal. Many organisations can fund experimentation before they can define the conditions under which experimentation becomes enterprise action.

The third is alchemy from chaos: the belief that AI can extract reliable operational intelligence from disordered organisational foundations. Firms take fragmented documents, inconsistent CRM records, unmanaged knowledge repositories, Slack histories, legacy systems, and poorly governed data lakes, then expect AI to produce coherent decision quality. That is not transformation. It is the outsourcing of unresolved management failure to a probabilistic system.

In Executive Intelligence Pipeline terms, alchemy from chaos is principally a Validation-stage failure: the organisation attempts to interpret and act on AI output before the underlying evidence base has been tested for quality, relevance, completeness, and fitness for decision.

Ryseff’s 2024 RAND analysis notes that, by some estimates, more than 80% of AI projects fail — twice the failure rate of non-AI IT projects. It identifies recurring root causes that reinforce this diagnosis: misunderstood problems, inadequate data, weak infrastructure, technology-first decision-making, and overambitious project selection. The executive lesson is clear: many AI failures begin before model selection. They begin when organisations do not know precisely which problem is being solved, what data is fit for purpose, how the workflow must change, and who owns the consequence.

The decision condition

A pilot is evidence only if it is designed to inform a decision. Too often, AI pilots are designed to create momentum rather than judgement. They show what might be possible in a contained environment, but they do not answer the harder enterprise questions: Can this be governed? Can it be scaled? Can it be audited? Can it be stopped? Can it be corrected when performance deteriorates?

Executives should not begin with the question, “Which AI tool should we buy?” They should begin with: “Which decision, process, or constraint must be improved, and what evidence would justify delegating part of that process to an AI-enabled system?”

That distinction is not semantic. It is governance.

A disciplined AI project begins with a decision condition: what must be decided, by whom, at what level of uncertainty, with what data, under what accountability, and with what consequences if the recommendation is wrong. Only then should the organisation determine whether AI is an appropriate instrument.

This is where the Decision Before Consequence framework becomes operationally useful. Developed and applied through professional work in security management, business continuity, defence innovation, and high-consequence decision analysis, the framework asks whether the organisation has achieved decision readiness before authorising consequential action. In AI settings, that means asking whether the enterprise can move from technological possibility to accountable deployment without losing control of evidence, authority, risk, and correction.

AI cannot carry consequence

The accountability problem is not theoretical. In Moffatt v. Air Canada, a Canadian tribunal held that Air Canada could be liable after its website chatbot gave a passenger misleading information about bereavement fares. The tribunal ordered compensation and rejected the practical effect of distancing the company from information supplied through its own automated customer interface. The monetary value was modest; the governance principle was not. A company cannot place an AI system between itself and the customer, then disown the consequence when the system misleads.

For executives, the case matters because it shows that AI systems may generate, classify, recommend, predict, or communicate, but they do not carry institutional responsibility. The organisation does. Where an AI system acts in the name of the enterprise, the enterprise remains accountable for the consequences.

This principle becomes more serious as AI moves from customer interaction into finance, employment, healthcare, security, logistics, and operational control. Reuters’ 9 February 2026 investigation, “As AI enters the operating room, reports arise of botched surgeries and misidentified body parts,” reported safety concerns involving AI-enabled medical systems, including adverse events, lawsuits, and evidence that AI-enabled devices had roughly twice the recall rate of comparable non-AI devices. The companies involved disputed causal interpretations in some cases, but the management lesson remains clear: adding AI to a complex operational system can increase risk when validation, oversight, and operating context are insufficiently mature.

This is where many board-level AI conversations remain underdeveloped. Executives often discuss AI in terms of productivity, labour substitution, speed, and innovation. They discuss risk as compliance, cybersecurity, or data privacy. But the deeper question is consequence ownership: who is responsible when an AI-supported decision causes financial loss, reputational harm, regulatory breach, unsafe action, or operational failure?

The answer cannot be: the model.

The Executive Intelligence Pipeline

The practical governance challenge is to convert AI capability into decision-relevant understanding before action is authorised. This is the purpose of the Executive Intelligence Pipeline:

Signal → Validation → Interpretation → Escalation → Decision → Action → Adaptation

AI can assist every stage of that pipeline. It can detect signals, structure data, identify anomalies, simulate options, classify risks, and recommend action. But it can also distort every stage if executives mistake output for understanding.

A signal is not intelligence. A prediction is not judgement. A recommendation is not a decision. Automation is not accountability.

The pipeline therefore gives executives a diagnostic question: where is the AI initiative failing? Is the signal weak? Is the data unvalidated? Is the interpretation wrong? Is escalation blocked? Is no accountable decision being made? Is action misaligned with authority? Is there no adaptation loop after deployment?

This matters because AI failure is rarely located only inside the model. It often appears in the spaces between data, workflow, authority, and consequence.

The NIST AI Risk Management Framework is useful precisely because it treats AI risk as something that must be governed, mapped, measured, and managed in context. AI governance is not a policy appendix. It is the executive discipline that determines whether AI can be converted from technical capacity into accountable organisational action.

Competitive AI adoption pressure

One reason organisations make weak AI decisions is that they are not acting only from evidence. They are acting under competitive pressure. Boards hear that competitors are adopting AI. Investors ask about AI strategy. Employees use public tools informally. Vendors frame delay as strategic backwardness. Executives fear being seen as slow.

The result is a dangerous institutional pattern: AI adoption becomes a signal of modernity rather than a disciplined instrument of strategy.

Gartner’s January 2026 forecast projected worldwide AI-specific spending of $2.52 trillion in 2026, a 44% year-on-year increase. The management question is not whether AI investment is increasing — it is whether executive decision systems are improving at the same rate.

High-discipline firms separate experimentation from deployment. They require a named decision owner. They define success criteria before pilots begin. They establish kill rules before sunk-cost bias accumulates. They test data readiness before model selection. They redesign workflows before automation. They ensure that escalation, override, audit, and correction mechanisms exist before scale.

Low-discipline firms do the opposite. They license horizontal AI tools broadly, encourage scattered experimentation, call adoption “transformation,” and search for financial impact afterwards. They permit pilots to proliferate without a clear route to production. They allow vendor roadmaps to substitute for internal judgement. They approve scaling before the organisation has determined how accountability, data integrity, workflow redesign, and human supervision will operate.

This is why horizontal AI often disappoints. General-purpose tools may help individuals summarise, draft, search, or analyse faster, but diffuse individual productivity does not automatically become enterprise-level profit. High-return AI tends to be more specific: tied to a high-friction workflow, a measurable decision bottleneck, a controlled data environment, and a clear operational owner.

Four executive moves: Diagnose, Map, Align, Commit

The remedy is not to slow AI adoption for its own sake. The remedy is to govern AI through decision readiness. Executives need a practical action set.

Diagnose the decision.
Begin with the decision, not the tool. Identify the business decision, operational bottleneck, risk condition, or workflow constraint the AI system is meant to improve. If the organisation cannot name the decision, it is not ready to scale the system.

Map the consequence.
Determine what happens if the system is wrong, incomplete, biased, outdated, overconfident, or misused. Consequence mapping must include financial, operational, legal, reputational, safety, and human effects. The point is not to eliminate risk; it is to know what risk is being authorised.

Align authority and accountability.
Every AI system that influences consequential action needs an accountable owner. Someone must have authority to approve, pause, override, correct, or terminate the system. If authority is distributed but accountability is vague, AI deployment will amplify organisational irresponsibility.

Commit with adaptation.
Scaling AI is not a one-time decision. It requires monitoring, feedback, escalation, and correction. Performance will drift. Data conditions will change. Users will adapt around the system. New risks will emerge. The organisation must therefore define kill rules, audit loops, and adaptation mechanisms before dependence forms.

These four moves convert AI from technology adoption into executive governance. They also distinguish disciplined transformation from strategy theatre.

The executive test

AI should not be judged by whether it appears impressive in demonstration. It should be judged by whether it improves a consequential decision or workflow under real organisational conditions.

The executive test is simple:

• Can the organisation specify the decision?

• Can it validate the evidence?

• Can it explain the model’s role?

• Can it assign accountability?

• Can it detect failure?

Can it reverse or correct the system before harm compounds?

If the answer is no, the organisation is not ready to scale. It may be ready to learn. It may be ready to experiment. But it is not ready to transform.

That distinction matters because the next phase of AI will be less forgiving than the first. Early experimentation tolerated ambiguity. Enterprise deployment will not. When AI moves into pricing, lending, hiring, procurement, medical decision support, fraud detection, infrastructure management, and strategic planning, consequences will become harder to contain.

AI can extend human sensing and speed, but it cannot carry consequence.

The organisations that succeed will not be those that spend the most visibly or adopt the fastest. They will be those that build the decision discipline required to convert AI capability into accountable action. The rest will continue to produce strategy theatre: impressive activity, weak governance, scattered pilots, rising costs, and limited strategic return.

The final test is not whether the organisation has an AI strategy. It is whether the organisation remains capable of judgement after AI enters the decision system. Without that discipline, the organisation is not leading the AI transition. It is financing consequences it has not yet learned how to govern.

Selected Sources and Evidence

MIT NANDA — Aditya Challapally, Chris Pease, Ramesh Raskar, and Pradyumna Chari, The GenAI Divide: State of AI in Business 2025, MIT Project NANDA, July 2025. The report states that despite $30–40 billion in enterprise generative AI investment, 95% of organisations were getting zero return, while only 5% of integrated pilots were extracting significant value. (Gartner)

Gartner CIO survey — Gartner, “Gartner Survey Finds All IT Work Will Involve AI by 2030; Organizations Must Navigate AI Readiness and Human Readiness to Find, Capture and Sustain Value,” 20 October 2025. The release reports that in a May 2025 survey of 506 CIOs and technology leaders, 72% said their organisations were breaking even or losing money on AI investments. (Gartner)

S&P Global Market Intelligence — S&P Global Market Intelligence, “AI experiences rapid adoption, but with mixed outcomes — highlights from VotE: AI & Machine Learning,” 30 May 2025. The report states that companies abandoning most AI initiatives before production rose from 17% to 42%, with organisations scrapping an average of 46% of proof-of-concept projects before broad adoption. (S&P Global)

IBM — IBM, “IBM Study: CEOs Double Down on AI While Navigating Enterprise Hurdles,” 6 May 2025. IBM reported that only 25% of AI initiatives had delivered expected ROI and only 16% had scaled enterprise-wide. (IBM Newsroom)

McKinsey & Company — McKinsey & Company, The State of AI: Global Survey 2025, 5 November 2025. McKinsey reported that 39% of respondents attributed any level of EBIT impact to AI, and most of those respondents said that less than 5% of their organisation’s EBIT was attributable to AI use. (McKinsey & Company)

RAND Corporation — James Ryseff, The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed, RAND Corporation, 2024; and RAND, “Why AI Projects Fail,” 10 April 2025. Ryseff’s analysis identifies root causes including misunderstood problems, poor data, inadequate infrastructure, technology-first decision-making, and overambitious project selection. RAND’s 2025 presentation notes that, by some estimates, more than 80% of AI projects fail — twice the failure rate of non-AI IT projects. (RAND Corporation)

Moffatt v. Air Canada — Moffatt v. Air Canada, 2024 BCCRT 149; see Bristows, “AI Chatbot flies solo and Air Canada foots the bill — Moffatt v Air Canada,” 27 March 2024. The case concerned misleading information supplied through Air Canada’s chatbot and illustrates that companies may remain responsible for automated customer-facing outputs.

Reuters medical-device investigation — Reuters, “As AI enters the operating room, reports arise of botched surgeries and misidentified body parts,” 9 February 2026. The investigation reported adverse events, lawsuits, and recall-rate concerns involving AI-enabled medical devices, while noting that manufacturers disputed causal interpretations in some cases. (Reuters)

NIST — National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1, U.S. Department of Commerce, January 2023. Canonical DOI: 10.6028/NIST.AI.100-1. The framework sets out a voluntary approach for organisations designing, developing, deploying, or using AI systems to manage AI risks through governance and the functions to map, measure, and manage AI risks. (NIST Publications)

Gartner AI-specific spending forecast — Gartner, “Gartner Says Worldwide AI Spending Will Total $2.5 Trillion in 2026,” 15 January 2026. Gartner forecast worldwide AI-specific spending of $2.52 trillion in 2026, a 44% year-on-year increase. This is distinct from Gartner’s separate total worldwide IT spending forecast, which includes broader categories such as data centre systems, devices, software, IT services, and communications services. (Gartner)

Author workflow disclosure

This article was produced through an AI-assisted but human-directed workflow. AI support was used for accessibility assistance, article structuring, language refinement, source-discovery prompts, revision planning, and conversion of editorial comments into specific amendments. The author retained responsibility for the argument, accepted or rejected suggested changes, checked the logic of the claims, and remained accountable for the final text. AI-generated material was not treated as empirical evidence, and synthetic or illustrative examples were not presented as observed data.

Image note

The image accompanying this article is AI-generated and is intended for illustration purposes only.

CONTACT

© 2026 Dr Danie Adendorff. All rights reserved. Rumbls.com is an independent analytical blog.

TERMS AND CONDITIONS

DISCLAIMER

COOKIE POLICY

PRIVACY POLICY