The Missing Half OF AI
Evidence

What the research actually says

Every figure this series uses, with its primary source. 18 anchors from MIT, BCG, PwC, McKinsey, Deloitte, KPMG, Gartner, RAND, Stanford, Harvard and EY. Where a number is a forecast rather than a measurement, or where the exact sentence isn’t on the page we link to, it says so.

Deep links were checked against the primary sources. Several publishers block automated access but load normally in a browser. Corrections welcome — tell me what I got wrong.

Value

The core. What the four dimensions below have to add up to.

77% vs 20%
77 percent versus 20 percent positive business outcomes, depending on how confident the organisation is in its talent.
KPMG Global AI Pulse Survey, 2026 · 20 countries, published 31 March 2026

Intent

Why, before the tool.

95%
95 percent of enterprise GenAI pilots show no measurable P&L impact
MIT (NANDA / Project NANDA), 2025
10-20-70
10 percent algorithm, 20 percent data and technology, 70 percent people, process and culture
BCG, 2025
80/20
80 percent of the value of an AI initiative comes from redesigning the work, 20 percent from the technology
PwC (2026 AI Business Predictions), 2026
#1 root cause
Miscommunication between the problem side and the solution side is the single most common root cause of failed AI initiatives — not the model, the data or the budget.
RAND Corporation, 2024
40% by 2027
Gartner predicts that more than 40 percent of agentic AI projects will be cancelled by the end of 2027.
Gartner, 2025
A prediction, not a measurement.

Trust

The real adoption curve.

66% / 46%
66 percent use AI regularly, but only 46 percent trust it.
KPMG / University of Melbourne (Trust, attitudes and use of AI), 2025 · n=48,340 across 47 countries, Nov 2024–Jan 2025
qualitative
Sustained AI use can erode trust within a team and the willingness to disagree — what the authors call trust ambiguity.
Jayshree Seth & Amy C. Edmondson, Harvard Business Review, 2026
qualitative
Trust in AI follows trust in leadership, not the accuracy of the tool.
David De Cremer, Harvard Business Review, 2025
71% (2023) / 64% (2025)
71 percent of employees report anxiety in connection with AI (2023); 64 percent of executives name fear of replacement, rather than augmentation, as a brake on AI adoption (2025).
EY — zwei getrennte Surveys, 2025
EY research, cited with attribution. Two separate surveys: the 71 percent are employees in 2023, the 64 percent are executives in 2025 speaking about adoption — not employees about their own use.
52%
52 percent say they feel nervous about AI.
Stanford AI Index 2026, 2026
Nervousness about AI products and services in general, not specifically at work.

Alliance

One team, not two halves.

50% by 2027
By 2027, 50% of companies that attributed headcount reduction to AI will rehire staff to perform similar functions, but under different job titles.
Gartner, 2026
A prediction, not a measurement. Issued by Gartner's Customer Service & Support practice; the forecast sentence itself names no industry limit, but the underlying data does. From the same source family, a figure that cuts the other way: only about 20 percent of customer service leaders reported any AI-driven headcount reduction.
+40% / -19pp
Within the capability frontier: +25% speed, +12% completion, +40% quality. Outside it: 19 points more often wrong.
Dell'Acqua, Mollick, Lakhani et al. (Harvard/BCG RCT), 758 Consultants, 2023
84%
84 percent of organisations have not redesigned work around AI
Deloitte State of AI in the Enterprise 2026, 2026
The figure is attributed to this edition; the exact sentence is not visible on the linked page.
+15% (QJE 2025)
Support agents gain 15 percent productivity on average. The distribution is the actual finding: the least experienced improve both speed and quality, while the most experienced gain little speed and lose some quality.
Brynjolfsson, Li, Raymond, Quarterly Journal of Economics, 2025
The widely cited 14% / 34% come from the 2023 NBER working paper. The version published in the Quarterly Journal of Economics reports 15 percent on average and does not put a number on novices in the abstract.

Leadership

What the machine can’t lead.

72% / 50%
72 percent of CEOs are now the primary decision-maker on AI, twice as many as the year before; 50 percent see job stability tied to AI strategy.
BCG AI Radar 2026, 2026 · n=2,360 including 640 CEOs, 16 markets
92% vs 62% (30 pts)
Where AI strategy is communicated clearly, 92 percent report a positive productivity effect — against 62 percent where communication is unclear.
EY Agentic AI Workplace Survey, 2025 · n=1,148 US desk workers, Aug–Sep 2025
EY research, cited with attribution.
28%
Only about 28 percent of organisations have CEO-level ownership of AI governance. Those are the cases that correlate most strongly with bottom-line impact, with workflow redesign the number one EBIT driver.
McKinsey (QuantumBlack, 'rewiring to capture value'), 2025