(20) Why AI Adoption Depends on Management

July 19, 2026/15 min read

Three studies, the same adoption gap

Three independent research programmes, using different methodologies and measuring different populations, arrived at nearly identical conclusions in 2025. MIT's Project NANDA found that 95 percent of enterprise generative AI pilots failed to deliver measurable financial impact, based on interviews with 150 executives, surveys of more than 350 employees, and analysis of 300 AI deployments. BCG's Build for the Future study of 1,250 organisations reported that only 5 percent qualified as "future-built" — generating substantial AI value at scale — while 60 percent were laggards with "little to show for their investment." McKinsey's Global AI Survey found that 88 percent of organisations now use AI in at least one function, but only 5.5 percent — the "AI high performers" — attributed more than 5 percent of EBIT to AI.

These numbers measure different failure modes. MIT captures pilots that never reach production; BCG captures maturity gaps across the adoption lifecycle; McKinsey captures financial insignificance. S&P Global Market Intelligence adds pipeline attrition: 42 percent of companies scrapped most AI initiatives in 2025, up from 17 percent the year before. Gartner forecast in mid-2024 that 30 percent of generative AI projects would be abandoned after proof of concept by the end of 2025, and by April 2026 it had extended the pattern to the next wave: more than 40 percent of agentic AI projects will be cancelled by the end of 2027, driven by escalating costs, unclear business value, and inadequate risk controls. Model limits account for some failures, but escalating cost, weak use-case selection, and inadequate control sit with management. By February 2026, according to Gartner data reported by Harvard Business Review, only about one in fifty AI investments was delivering transformational value. Most enterprises remain stuck in what McKinsey calls “activity without outcomes.”

What the five percent do differently

The firms generating material returns from AI share organisational characteristics that have nothing to do with which large language model they use. BCG reports that nearly 100 percent of future-built organisations have deeply engaged C-suites directing AI strategy, compared with just 8 percent of laggards, translating business goals into a multiyear, fully funded vision rather than launching scattered pilots. McKinsey's survey found that workflow redesign had the biggest effect — out of 25 attributes tested — on an organisation's ability to see EBIT impact from generative AI. KPMG's Global AI Pulse for the second quarter of 2026, drawn from more than 2,100 senior technology decision-makers across 20 countries, put the sharpest number on this mechanism: organisations with clearly defined accountability for AI outcomes established ROI at three times the rate of those without — and sponsoring an implementation, KPMG stressed, is not the same as owning accountability for it.

Governance functions as a scaling mechanism, not overhead. More than 60 percent of BCG's future-built firms rigorously track AI value, compared with 17 percent of stagnating companies — telling them which deployments to expand and which to kill. Deloitte's 2026 State of AI report delivers the starkest data on the integration gap: 84 percent of surveyed companies have not redesigned jobs around AI capabilities even while rapidly expanding access to tools, and the top talent response is education at 53 percent, not role or workflow redesign. Gartner reports that 78 percent of CHROs agree workflows must change, yet only just over half of organisations have actually redesigned roles — even though its data shows firms are twice as likely to exceed revenue goals when they redesign work.

The Academic Evidence for Organisational Co-Production

Field experiments have begun to map the precise boundary between where AI works autonomously and where it requires organisational support. Brynjolfsson, Li, and Raymond published a field experiment in the Quarterly Journal of Economics in 2025 studying 5,172 customer support agents at a Fortune 500 firm. AI assistance increased productivity by 15 percent, with gains accruing disproportionately to less experienced workers, who saw a 34 percent improvement, while top performers saw minimal change — the AI effectively disseminated best practices to lower performers. The insight that matters: this success occurred in a well-defined, instrumented workflow — precisely the kind of environment that management creates.

A separate field experiment by Stanton and colleagues, published as NBER Working Paper 33795 in 2025, studied 7,137 knowledge workers across 66 firms. Workers with AI access spent two fewer hours per week on email, but the study found no shift in the composition of their tasks — distinguishing changes workers can make independently, like drafting emails faster, from changes requiring cross-team coordination, like restructuring meetings or reassigning responsibilities. That coordination boundary is a management challenge, not a model-quality challenge. Dell'Acqua and colleagues at Harvard Business School, in a field experiment with 758 BCG consultants, found that for tasks within AI's capability frontier consultants using AI were 12.2 percent more productive and produced more than 40 percent higher quality work, but for a task outside the frontier they were 19 percentage points less likely to produce correct solutions. Knowing where the frontier lies is a management judgment, not a technical specification.

These findings build on a longer literature. Bresnahan, Brynjolfsson, and Hitt established in 2002 that information technology, workplace organisation, and innovation are complements — technology value is "co-invented" through organisational redesign, not simply installed. Brynjolfsson, Rock, and Syverson extended this in their 2021 "Productivity J-Curve" paper: general-purpose technologies require intangible complementary investments poorly measured in the early adoption period. This explains why macro-level AI productivity data looks disappointing. The San Francisco Federal Reserve noted in February 2026 that "most macro-studies of productivity growth find limited evidence of a significant AI effect," a CEPR survey of more than 5,000 executives found 89 percent reported no productivity impact over the prior three years, and Nobel laureate Daron Acemoglu held in a June 2026 assessment to his estimate of roughly 0.55 percent in total-factor-productivity gains over the coming decade, warning that agentic AI could accelerate the effect only if smaller firms build the capacity to adopt it. That gap between expectation and evidence is the macro-level expression of the management deficit.

Automation or Augmentation Is a Management Choice

The clearest new evidence that organisational choices — not model capability — determine outcomes arrived in mid-2026 from the Stanford Digital Economy Lab and ADP Research, whose Canaries Dashboard tracks 4.6 million workers across 730-plus occupations. In the most AI-exposed occupations, employment for workers aged 22 to 25 was contracting 3.8 percent year over year, while the least-exposed roles in the same age group grew about 2 percent, even as the aggregate effect across all ages stayed muted. Erik Brynjolfsson attributed the entry-level decline to automation rather than augmentation of junior tasks, warning it could widen inequality absent job redesign. Whether AI displaces workers or complements them is not a property of the technology; it is a consequence of how management chooses to redesign the work — the same variable that separates the value-creating 5 percent from everyone else.

Shadow AI exposes the management gap

The most operationally revealing indicator of management lag is not model hallucination or integration complexity. It is unsanctioned adoption. Microsoft and LinkedIn's Work Trend Index reported in 2024 that 78 percent of AI users bring their own tools to work, and MIT's 2025 research found that employees at more than 90 percent of companies use personal AI accounts while only 40 percent of organisations provide official large language model tools. The pattern has only intensified in 2026: a Smarsh study in July found shadow AI "outpacing enterprise governance," and a Synyega analysis in May named it the fastest-growing component of software sprawl. A March 2026 survey by the Purple Book Community and ArmorCode found that 59 percent of security leaders confirm or suspect employees use unapproved AI tools, and 73 percent say AI-assisted development is outpacing security review capacity.

Shadow AI is a security problem produced in part by a management gap. When organisations do not provide strategic direction, sanctioned tools, and governed pathways, employees route around them.

Governance is becoming a hard constraint

Governance has shifted from best practice to legal obligation — and the regulatory response has itself become evidence for the thesis. The EU AI Act, Regulation 2024/1689, established a risk-based compliance regime with fines reaching 7 percent of global turnover for the most serious violations. Yet on 29 June 2026 the Council of the EU gave final approval to the Digital Omnibus on AI, deferring the stand-alone high-risk obligations under Annex III from 2 August 2026 to 2 December 2027 and those for AI embedded in regulated products to 2 August 2028, while transparency and general-purpose AI rules stayed on schedule. The stated reason was that the technical standards firms need to comply do not yet exist — not a regulatory retreat, but an admission that the governance infrastructure required to operationalise AI lags its deployment. The clock moved. The exposure did not.

Financial regulators are pushing in the same direction, straight at the boardroom. In April 2026 the Federal Reserve, OCC, and FDIC issued guidance under SR 26-2 that deliberately moved generative and agentic AI out of the formal Model Risk Management scope while reminding banks they still own the governance, monitoring, and "effective challenge" for those systems — refusing to write technology-specific rules and instead assigning accountability to board governance, the clearest possible statement of "management, not technology" from a tier-one supervisor. By mid-2026 a professional consensus, spanning KPMG and INSEAD's governance principles, the NACD's board AI governance guide, and Reuters Practical Law, had settled that AI oversight now falls squarely within directors' existing duty of care. What was an inference in March had become a fiduciary expectation by July.

The gap between deployment ambition and governance readiness remains the most dangerous data point in the landscape. Deloitte's 2026 data shows that 74 percent of companies plan to deploy agentic AI within two years, but only 21 percent report a mature governance model for autonomous agents. Gartner projects that 40 percent of enterprise applications will feature task-specific AI agents by the end of 2026, up from under 5 percent in 2025, even as it forecasts that more than 40 percent of agentic projects will be cancelled by 2027; PwC found only about 11 percent of enterprises have agents actually running against real business workflows in production. That distance between ambition and control is the operational definition of management debt.

Some of the technology is not ready

The most credible objection to the management thesis is that the technology genuinely is not ready. BCG's survey data includes hallucinations, model reliability, integration complexity, and scaling costs among executive-reported challenges; Dell'Acqua's "jagged frontier" and RAND's study of AI project failures point to real technical limits. These are engineering constraints, not management failures — but they do not explain the aggregate pattern.

Most firms failing to extract value from AI are not attempting frontier applications in clinical oncology or autonomous trading. They are struggling with email summarisation, customer service augmentation, and document drafting — tasks well within AI's demonstrated capability. The 95 percent figure reflects organisations that cannot define which problems to solve, govern how tools are deployed, or redesign workflows to integrate AI outputs. The technology works in instrumented, well-managed environments, and management is what creates them.

The pattern predates AI

The pattern is not new. According to industry analyses, ERP implementations in the 1990s and 2000s produced failure rates of 55 to 75 percent, with only 23 percent considered successful. The causes were structurally identical: technology treated as a silver bullet, insufficient change management, and lack of executive sponsorship. Research on that era concluded that successful ERP requires treating it not as a technology project but as an organisational transformation initiative. Cloud migration in the 2010s followed the same arc, bottlenecked by the "lift and shift" mentality that moved workflows unchanged into new infrastructure, and Six Sigma and lean initiatives failed at similar rates when treated as technical exercises. McKinsey's finding that roughly 70 percent of change programmes fail describes a management pattern, not a technology one. The lesson is structural: general-purpose capabilities do not create value until organisations redesign themselves to use them.

Management determines the return

BCG reports that future-built firms achieve five times the revenue increases and three times the cost reductions of laggards; KPMG finds that organisations with clear accountability establish ROI at three times the rate of those without it. Meanwhile, enterprise AI spending is projected at 2.59 trillion dollars in 2026, with AI agent software reaching 206.5 billion dollars, up 139 percent year over year. Spending is accelerating while cancellation rates remain high. The model matters, but so does the operating system around it: accountable ownership, workflow redesign, measurement, and control. AI adoption is an organisational redesign decision as much as a technology decision.

Sources

AI Failure Rates and Value Gap

MIT NANDA, "The GenAI Divide: State of AI in Business 2025" https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/
BCG, "The Widening AI Value Gap — Build for the Future 2025 Global Study" https://www.bcg.com/publications/2025/are-you-generating-value-from-ai-the-widening-gap
McKinsey & Company, "The State of AI: Global Survey 2025" https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
Gartner, "30% of GenAI Projects Abandoned After POC by End of 2025" https://www.gartner.com/en/newsroom/press-releases/2024-07-29-gartner-predicts-30-percent-of-generative-ai-projects-will-be-abandoned-after-proof-of-concept-by-end-of-2025
Gartner, "Over 40% of Agentic AI Projects Will Be Canceled by End of 2027" https://www.digitalapplied.com/blog/agentic-ai-project-cancellations-gartner-40-percent-2026
S&P Global Market Intelligence, "AI Project Failure Data" https://www.ciodive.com/news/AI-project-fail-data-SPGlobal/742590/
Deloitte AI Institute, "State of AI in the Enterprise 2026" https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html
RAND Corporation, "AI Projects Fail at Twice the Rate of IT Projects" https://www.rand.org/pubs/research_reports/RRA2680-1.html
Aykens et al., "9 Trends Shaping Work in 2026 and Beyond" https://hbr.org/2026/02/9-trends-shaping-work-in-2026-and-beyond

Characteristics of AI Value Creators

BCG, "The Widening AI Value Gap — Full Report" https://media-publications.bcg.com/The-Widening-AI-Value-Gap-Sept-2025.pdf
KPMG, "Global AI Pulse Q2 2026" https://kpmg.com/us/en/media/news/q2-ai-pulse-2026.html
CIO Dive, "AI Confidence Is on the Rise Despite Cost Concerns" https://www.ciodive.com/news/ai-confidence-rise-despite-cost/824738/
IDC/NetApp, "Scaling Enterprise AI Responsibly" https://www.netapp.com/media/142474-idc-2025-ai-maturity-findings.pdf
Gartner, "Top Change Management Trends for CHROs in Age of AI" https://www.gartner.com/en/newsroom/press-releases/2026-3-16-gartner-identifies-top-change-management-trends-for-chros-in-age-of-ai
Deloitte, "State of AI in the Enterprise 2026 — Global Report" https://www.deloitte.com/content/dam/assets-shared/docs/about/2025/state-of-ai-2026-global.pdf
IDC/Microsoft, "Frontier Firms: Unlocking the Business Value of AI" https://www.microsoft.com/en-us/microsoft-cloud/blog/2025/12/11/becoming-a-frontier-firm-unlocking-the-business-value-of-ai/
Microsoft/LinkedIn, "Work Trend Index 2024" https://www.microsoft.com/en-us/worklab/work-trend-index/
Li, Zhu, and Hua, "Overcoming the Organizational Barriers to AI Adoption" https://hbr.org/2025/11/overcoming-the-organizational-barriers-to-ai-adoption

Academic Field Experiments and Economic Theory

Brynjolfsson, Li, and Raymond, "Generative AI at Work" https://academic.oup.com/qje/article/140/2/889/7990658
Stanton et al., "Shifting Work Patterns with Generative AI" https://www.nber.org/papers/w33795
Dell'Acqua et al., "Navigating the Jagged Technological Frontier" https://www.hbs.edu/faculty/Pages/item.aspx?num=64700
Dell'Acqua et al., "The Cybernetic Teammate" https://d3.harvard.edu/the-cybernetic-teammate-how-ai-is-reshaping-collaboration-and-expertise-in-the-workplace/
Bresnahan, Brynjolfsson, and Hitt, "Information Technology, Workplace Organization, and the Demand for Skilled Labor" https://academic.oup.com/qje/article-abstract/117/1/339/1851770
Brynjolfsson, Rock, and Syverson, "The Productivity J-Curve" https://www.aeaweb.org/articles?id=10.1257/mac.20180386
Gambacorta et al., "Generative AI and Labour Productivity: A Field Experiment on Coding" https://www.bis.org/publ/work1208.pdf

Historical Parallels and Change Management

Panorama Consulting, "ERP Failure in the 90s vs Today" https://panorama-consulting.com/erp-failure-in-the-90s-vs-today/
Air Academy Associates, "The Psychology of Change: Why 70% of Six Sigma Projects Fail" https://airacad.com/psychology-of-change-why-70-six-sigma-projects-fail/
Vest and Gamm, "A Critical Review of Healthcare Transformation Strategies" https://pmc.ncbi.nlm.nih.gov/articles/PMC2709888/

Industry and Market Data

Gartner via BERI, "AI Agent Adoption and Software Spending 2026" https://www.beri.net/article/ai-agent-adoption-enterprise-2026-gartner-idc
Kurums, "Why Are 40% of Agentic AI Projects Being Canceled Despite Record Enterprise Spending in 2026" https://kurums.com/why-are-40-of-agentic-ai-projects-being-canceled-despite-record-enterprise-spending-in-2026/