The 6% Question
Researched by an agentic pipeline · reviewed and gated by the author
AI spending without workflow redesign is, for most enterprises, a consumption expense, not a capital investment.
The Conversion Failure
Four independent surveys of more than 12,000 executives, conducted across 2025 and 2026, converge on a single finding: the vast majority of enterprises spending on AI cannot attribute any of that spending to earnings. McKinsey’s 2026 global survey — 1,719 respondents across 97 nations, GDP-weighted — reports that 37 per cent of organisations attribute any EBIT impact to AI, unchanged from 2025 despite adoption scaling from 38 per cent to 44 per cent [S1]. Only 6 per cent attribute 5 per cent or more of EBIT to AI. That figure, too, is flat year on year.
This is not an outlier result. The NBER’s multi-country study of nearly 6,000 executives, conducted by the Federal Reserve Bank of Atlanta, the Bank of England, the Deutsche Bundesbank and Macquarie University, found that 89 per cent reported zero labour productivity impact from AI over three years [S3]. KPMG’s Q2 2026 survey of 2,145 C-suite leaders across 20 markets found 7 per cent reporting established ROI [S5]. BCG’s 2026 AI Radar found 5 to 8 per cent at scale [S7]. The convergence across methodologies, geographies and sample frames lifts this finding above any single study’s limitations.
The Perception-Measurement Disconnect
Eighty per cent of AI users in McKinsey’s survey report that AI has improved their individual productivity. Fifty per cent report better decisions. Only 13 per cent feel anxious about their career prospects [S1]. By every subjective measure, AI is working.
The UK government’s trial of Microsoft 365 Copilot puts this perception under controlled light. Three hundred civil servants used Copilot for three months. Seventy-two per cent expressed satisfaction. The Net Promoter Score was 31 — rated good. Participants were disappointed when the trial ended [S6].
Measured productivity gain: zero.
PowerPoint slides took seven minutes longer on average and required corrections for inferior quality. Excel analysis was slower and less accurate. The trial found “no conclusive evidence that these led to measurable productivity improvement” [S6]. Individual satisfaction is real. Measured enterprise output gain is not.
What the 6 Per Cent Did Differently
McKinsey’s high performers — the 6 per cent attributing 5 per cent or more of EBIT to AI — are not distinguished by their technology choices or their vendor selections. They are distinguished by organisational redesign [S1][S2].
Nearly three-quarters (73 per cent) of high performers have fundamentally redesigned workflows around AI, up from 55 per cent in 2025. Among all other respondents, that figure is 25 per cent. High performers are 3.3 times more likely to be planning fundamental business transformation within three years. They spend more — more than twice as likely to allocate over 15 per cent of their IT budget to AI — but the spending follows the redesign, not the reverse.
The technology is a necessary condition; the organisational change is the sufficient one.
The Spending Reality
Enterprise AI budgets are substantial and growing. KPMG reports an average of $188 million in Q2 2026, consistent with $186 million in Q1 [S5]. Twenty-eight per cent of enterprises spend more than 10 per cent of their IT budget on AI. Sixty per cent plan to increase investment next year. Only 6 per cent of McKinsey’s respondents would cut spending if initiatives underdelivered [S2].
The allocation within those budgets is poorly matched to evidence. More than half of AI spending concentrates in sales and marketing, where measurability is lowest, while documented ROI cases — customer service automation at Klarna ($39 million saved), 1-800Accountant (90 per cent case deflection), OpenTable (70 per cent autonomous resolution) — sit in service functions that receive less investment [S7]. MIT’s research found that 42 per cent of firms discontinued most AI projects in 2025 rather than continuing iteration, and 95 per cent of generative AI pilots showed zero measurable profit-and-loss impact [S7].
What Remains Unproven
All EBIT attribution in these surveys is self-reported. No study independently verified financial statements against AI spending. The “high performer” classification relies on executives’ own assessment of their organisations’ AI-to-earnings conversion [S1].
McKinsey’s conflict of interest is material: the survey identifies organisational transformation as the remedy for the conversion failure, and McKinsey sells organisational transformation consulting. This does not invalidate the finding — it converges with independent evidence — but it should temper the prescription.
The causal mechanism linking workflow redesign to EBIT remains correlational. High performers may share other characteristics — management quality, change-readiness, sector positioning — that independently drive both redesign and earnings growth. The 80 per cent individual productivity figure is perception, not measurement: the UK Copilot trial demonstrates that satisfaction and measured output can move in opposite directions [S6]. Forward workforce predictions have historically overshot by a factor of two [S1].
What a Portfolio Decision-Maker Should Conclude
The question is not whether to invest in AI. It is whether your organisation has the change capacity to convert that investment into earnings — and whether your governance framework can distinguish the two.
Three conclusions survive the evidence.
First, AI spending without workflow redesign is, for most enterprises, a consumption expense, not a capital investment. The 94 per cent are not underspending on technology. They are underspending on the organisational change that makes technology productive. Investment committees should require evidence of workflow redesign — not adoption metrics — before approving further AI capital allocation.
Second, self-reported productivity gains are not evidence of enterprise value. Four independent studies confirm that individual perception diverges from organisational measurement. Governance frameworks that accept user satisfaction as ROI evidence are measuring the wrong thing.
Third, if your AI programme cannot demonstrate earnings impact within two budget cycles, the constraint is almost certainly organisational, not technological. Additional technology spending will not resolve it. The capital is better directed at the specific practice that distinguishes the 6 per cent — fundamental workflow redesign — or reallocated to the functions where documented ROI already exists.
Sources
- McKinsey QuantumBlack — The State of AI: Global Survey 2026 — August 2026 — https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
- TechTimes — Record AI Spending Can’t Move Earnings Needle for 94% of Enterprises, McKinsey Finds — August 2026 — https://www.techtimes.com/articles/325590/20260826/record-ai-spending-cant-move-earnings-needle-94-enterprises-mckinsey-finds.htm
- NBER — Firm Data on AI (Working Paper 34836) — February 2026 — https://www.nber.org/papers/w34836
- NBER Digest — Global Evidence on Business Use of AI — May 2026 — https://www.nber.org/digest/202605/global-evidence-business-use-ai
- KPMG — Global AI Pulse Q2 2026 — June 2026 — https://kpmg.com/xx/en/media/press-releases/2026/06/growing-adoption-signals-progress-as-cost-visibility-and-accountability-drive-ai-value.html
- Computing.co.uk — UK Government Trial of Microsoft 365 Copilot Reveals No Clear Productivity Boost — September 2025 — https://www.computing.co.uk/news/2025/uk-government-trial-of-microsoft-365-copilot-reveals-no-clear-productivity-boost
- Value Add VC — Enterprise AI ROI Measurement in 2026 — 2026 — https://valueaddvc.com/blog/enterprise-ai-roi-in-2026-what-companies-are-actually-measuring-and-finding