The Redesign Deficit

The Edge · Transformation · Wednesday, 7 October 2026

Researched by an agentic pipeline · reviewed and gated by the author

The J-curve tells organisations to wait. The redesign evidence tells them what to do while they wait.

The Dashboard Paradox

Every indicator on the transformation dashboard says AI is working. Adoption is at 87 per cent. Individual productivity scores are strong. Usage is rising month on month. The programme team reports enthusiasm. The board paper shows investment on track.

Then the CFO presents: EBIT contribution from AI is flat. No measurable financial return. Same as last year.

This is not a hypothetical. It is the documented condition of the vast majority of enterprises deploying AI in 2026. And the evidence for it is no longer anecdotal — five independent surveys, conducted across different populations, geographies, and methodologies, have converged on the same quantitative finding within months of each other. Individual productivity gains from AI are large and real, yet they are systematically failing to appear in organisational financial performance. The question is not whether this is happening. It is why.

What the Surveys Found

McKinsey’s State of AI survey, covering 1,719 respondents across 97 nations, reports that 80 per cent of AI users experience individual productivity gains — yet only 37 per cent of organisations attribute any EBIT impact to AI, unchanged from 2025 [S1]. Deloitte’s global survey of 3,235 leaders finds 66 per cent achieving productivity improvements but only 20 per cent generating revenue growth [S3]. The CEPR/Duke CFO Survey, drawing on 734 senior financial executives, measures implied labour productivity growth of just 0.6 per cent in 2025, expected to reach 1.8 per cent in 2026 — but notes that firms consistently report larger productivity gains than those actually measured in revenue and employment changes [S4].

The convergence is the finding. Three large, independent, methodologically different surveys arriving at the same structural conclusion within months: individual gains are not translating into organisational value. The gap is not marginal. It is the dominant condition.

What makes this evidence particularly consequential is the consistency of its explanation. These surveys do not merely document the gap. They identify the same differentiating variable.

Where the Gains Go

The most precise evidence for the dissipation mechanism comes from outside the enterprise survey tradition. Humlum and Vestergaard’s NBER study, drawing on Danish administrative panel data linked to workplace surveys, documents what happens to AI-generated productivity at the individual level: nothing measurable [S5]. Despite 93 per cent adoption in supported workplaces, the researchers find precise null effects on worker earnings and hours — ruling out effects larger than 2 per cent two years after deployment. The explanation is instructive: 85 per cent of AI users reallocate time savings to other job tasks. The efficiency gain is real. It is also invisible to every organisational metric that matters.

This pattern operates through three interconnected channels.

The first is task-level gain without workflow integration. An employee produces faster work that enters the same approval queue, reaches the same reviewer on the same review cadence, and moves through the same operational cycle. The individual gained hours. The process captured none of them.

The second is time reallocation without output measurement. The NBER finding is critical here: saved time does not evaporate. Workers use it — they take on adjacent tasks, respond to more requests, produce more intermediate work. But without redesigned measurement systems, this activity is invisible. The worker is demonstrably busier. Organisational output, as measured, has not moved.

The third is governance overhead and workforce friction. Rapid deployment without structural preparation generates new costs. Writer’s survey of 2,400 respondents reports that 55 per cent of executives describe AI usage as a chaotic free-for-all, while 29 per cent of employees acknowledge actively undermining their organisation’s AI strategy [S2]. These figures carry a commercial caveat — the survey’s sponsor is an AI vendor, and the definition of sabotage is broad — but the friction they describe is corroborated by Deloitte’s finding that 37 per cent of organisations use AI superficially with minimal process changes [S3]. Deployment without redesign does not merely fail to produce value. It generates new work — compliance, oversight, damage control — that absorbs some of the individual productivity it enables.

Consider a composite that illustrates the pattern. A professional services firm deploys AI-assisted research and drafting across its advisory teams. Individual consultants report saving eight to ten hours per week. The firm’s utilisation metrics remain flat. Revenue per consultant is unchanged. The consultants used their time savings to take on more internal work — contributing to proposals, attending practice development sessions, responding to knowledge-sharing requests. Each activity is legitimate. None of it is billable. The firm’s measurement systems track billable utilisation, not total productive output. The AI gain was real, absorbed, and unmeasured. The CFO saw nothing.

This is not a technology failure. It is an architecture failure. The organisation deployed AI into an operating model designed to measure something the AI does not directly change.

The Honest Counter-Argument

There is a credible alternative explanation, and intellectual honesty requires stating it plainly. The flat financial returns may reflect the trough of a normal technology adoption curve, not a permanent structural deficiency.

The historical precedent is directly relevant. Robert Solow’s 1987 observation — that computers were visible everywhere except the productivity statistics — described a paradox that persisted for roughly a decade before information technology produced measurable economy-wide gains. Brynjolfsson and colleagues have argued persuasively that general-purpose technologies follow a J-curve: early deployment depresses measured productivity as organisations invest in complementary changes, before those investments generate returns.

The NBER evidence is the strongest support for this interpretation. The genuine task reorganisation it documents — 42 per cent of new AI-related tasks involving content generation, 35 per cent dedicated to quality review and compliance — is exactly the kind of complementary investment the J-curve model predicts [S5]. Workers and employers are restructuring, just not yet visibly. Under this reading, patience rather than redesign is the appropriate response.

The counter-argument deserves weight. But it does not eliminate the actionable core of the evidence. The J-curve predicts that gains will eventually materialise across the economy. It does not explain why, today, 6 per cent of organisations are already capturing measurable financial value while 94 per cent are not. If the gap were purely a matter of timing, early movers and late adopters should differ by deployment date or technology sophistication. They do not. They differ by something else entirely.

What Separates the Six Per Cent

McKinsey’s data identifies the single most discriminating variable [S1]. Among the 6 per cent of organisations qualifying as AI high performers — those reporting five per cent or more of EBIT from AI — 73 per cent have fundamentally redesigned workflows around AI capabilities. Among all other organisations, the figure is 25 per cent. This is a threefold architectural gap, and it has widened: in 2025, the high-performer redesign rate was 55 per cent.

The nature of the redesign matters. High performers did not simply adopt more aggressively or choose better tools. They changed what the organisation measures, who makes decisions with AI-generated inputs, and how workflows absorb individual productivity gains into collective output. Deloitte’s data confirms the bifurcation independently: only 34 per cent of organisations are using AI to deeply transform operations, while 37 per cent apply it superficially to existing processes [S3]. The California Management Review, synthesising cross-sector evidence, finds that only 30 per cent of AI pilots transition to scaled impact — and identifies workflow redesign as the critical enabler of that transition [S7].

A necessary caution: the redesign differential is correlational. Organisations capable of fundamental workflow redesign may already possess the managerial capability, governance structures, and organisational capital that enable them to capture value from any technology investment. The causal direction — whether redesign produces value or whether capable organisations both redesign and capture value — has not been established experimentally. The strongest honest reading is that redesign is necessary but not certainly sufficient: its absence reliably predicts failure to capture value, even if its presence does not guarantee success.

A further measurement caveat must be preserved. All of the productivity figures in this analysis rest partly on self-reported survey data. METR’s study of 349 technical workers found that people overestimate AI’s effect on their task time by an average of 40 percentage points [S6]. The individual-to-organisational gap is robust — it appears in administrative data as well as surveys — but the precise magnitude of individual gains likely carries significant upward bias.

The Governance Test

The most consequential implication of this evidence is not that organisations need to redesign their workflows, though many do. It is that the metrics governing AI investment decisions are systematically misleading.

Boards and transformation committees measuring adoption rates, individual productivity improvements, and deployment velocity are tracking indicators that do not predict organisational value capture. The evidence is unambiguous on this point: adoption and individual productivity can be high — 80 per cent high — while financial return is flat. These are not leading indicators that will eventually close the gap. They are measures of a different thing entirely.

The diagnostic question for any organisation investing in AI is not how much it has deployed but how much it has redesigned. Specifically: has any workflow been restructured so that AI-generated efficiency flows to a measured organisational output rather than being absorbed by task expansion? Has any decision-rights allocation changed because AI now produces inputs that were previously unavailable? Has the measurement system been adjusted to capture the value that AI-augmented employees actually create?

If the answer to all three is no, the absence of financial return is not a mystery. It is an architectural prediction.

Whether the broader productivity J-curve eventually delivers economy-wide gains is a question that only time can answer. The NBER evidence suggests genuine reorganisation is underway beneath flat aggregate statistics, and two to three years of enterprise AI deployment may be too short for definitive structural conclusions. But the J-curve tells organisations to wait. The redesign evidence tells them what to do while they wait. For transformation leaders, the distinction between these two responses is the difference between a governance posture and a governance programme.

The 6 per cent are not waiting for the curve to bend. They are bending it.

Sources

1. McKinsey & Company / QuantumBlack — The State of AI: Global Survey 2026 — August 2026 — https://www.mckinsey.com.br/capabilities/quantumblack/our-insights/the-state-of-ai
2. Writer / Workplace Intelligence — Enterprise AI Adoption in 2026 — April 2026 — https://writer.com/blog/enterprise-ai-adoption-2026/
3. Deloitte — The State of AI in the Enterprise 2026 — 2026 — https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html
4. CEPR / Duke University — AI, Productivity, and Work: Evidence from US Firms — 2026 — https://cepr.org/voxeu/columns/ai-productivity-and-work-evidence-us-firms
5. Anders Humlum and Emilie Vestergaard — Still Waters, Rapid Currents — NBER Working Paper 33777 — March 2026 — https://www.nber.org/system/files/working_papers/w33777/w33777.pdf
6. METR — Measuring the Self-Reported Impact of Early-2026 AI on Technical Worker Productivity — May 2026 — https://metr.org/blog/2026-05-11-ai-usage-survey/
7. California Management Review — Bridging the Gaps in AI Transformation — November 2025 — https://cmr.berkeley.edu/2025/11/bridging-the-gaps-in-ai-transformation-an-evidence-based-framework-for-scalable-adoption/