Five consulting firms. Five independent research programmes. Five separate datasets, methodologies, and sample populations. And the same conclusion: enterprise AI adoption has reached unprecedented levels, but the financial return on that adoption remains concentrated in a small minority of organisations. The rest are spending more, deploying more, reporting more activity — and seeing far less on the income statement than they expected.
This article draws on the published findings from McKinsey's State of AI 2025, BCG's "The Widening AI Value Gap" (Build for the Future 2025), Accenture's reinvention research, Deloitte's State of Generative AI in the Enterprise, and Bain's enterprise surveys. Each approaches the question from a different angle. Read together, they point unambiguously at the same root cause — and it is not how much you spend.
The scale of the gap
McKinsey separates usage from impact. The McKinsey Global Institute's estimate that generative AI could add the equivalent of $2.6 to $4.4 trillion annually across 63 use cases remains the headline figure for what the technology could deliver. McKinsey's 2025 State of AI survey shows what it currently does: 88% of organisations now use AI in at least one function, up from 78% a year earlier, but only about 39% report any EBIT impact at the enterprise level — and most of those put that impact at less than 5% of EBIT. The group McKinsey calls AI high performers — organisations attributing 5% or more of EBIT to AI and reporting significant value — is roughly 6% of respondents. Everyone else has the tool in the building without the result on the P&L. The full McKinsey analysis examines this gap in detail.
BCG draws the same line with different data. In BCG's "Build for the Future" research, only 5% of companies qualify as "future-built" — BCG's term for firms that systematically build AI capability across functions and consistently generate value from it. The performance differential is stark: compared with laggards, future-built firms show 1.7 times the revenue growth, 3.6 times the three-year total shareholder return, and 1.6 times the EBIT margin. These are not incremental advantages; they describe a different trajectory. The BCG maturity blueprint breaks down the capability architecture that separates these tiers.
Accenture measures the acceleration. Accenture's Reinventors — companies that have fundamentally reorganised operations around technology rather than bolting tools onto them — saw revenue growth roughly 15 percentage points higher than the rest of its sample between 2019 and 2022, a gap Accenture projects will widen to around 37 percentage points by 2026. Their margin advantage was real too: about 5.6 percentage points higher profitability over the same window. The Reinventors' edge is not that they spend more. It is that their spending compounds, because it sits on top of structural change rather than unchanged processes. The reinvention premium analysis explores what distinguishes Reinventors from the rest.
Deloitte captures the paradox of progress. Deloitte's State of Generative AI in the Enterprise — drawn from thousands of AI-literate leaders across more than a dozen countries — documents genuine forward motion: two-thirds of organisations report productivity and efficiency gains, and the share of companies with 40% or more of their projects in production is set to climb sharply. But the same research exposes the catch. The single biggest barrier respondents name is the AI skills gap, and only about one in five organisations reports a mature model for governing autonomous AI agents. More AI in production is not the same as more AI creating value — and Deloitte's own data shows the operational maturity required to convert one into the other lagging well behind the deployment curve.
Bain documents breadth without depth. Bain reports that roughly 95% of US companies now use generative AI, up about 12 percentage points in a year, with the average number of use cases in production doubling between late 2023 and the end of 2024. The expansion is broad but, for most, shallow. Bain finds that the firms actually capturing 10 to 25% EBITDA gains are tech-forward enterprises that broke through the pilot phase by redesigning high-value workflows end to end — not by accumulating more use cases. The majority running generative AI across many functions see real individual-level productivity that never aggregates into enterprise financial outcomes.
The five-study convergence
The common denominator is not technology spending. If it were, the gap would be closing as budgets rise. They are rising — Bain puts average annual AI budgets at roughly double their early-2024 level — and the gap is not closing. The common denominator is structural transformation, and all five studies describe it in strikingly similar terms despite their different frameworks.
Workflow redesign, not tool deployment. This is the sharpest finding in the entire corpus. McKinsey identifies the fundamental redesign of workflows as the factor with the single biggest effect on whether an organisation sees EBIT impact from generative AI — and high performers are nearly three times as likely to have done it (around 55% versus 20% of everyone else). BCG's future-built firms have re-architected their operating models. Accenture's Reinventors reorganised operations. Bain names end-to-end workflow redesign as the prerequisite for EBITDA impact. The vocabulary differs across four firms. The finding is identical: AI layered onto an unchanged process produces an unchanged result.
Business ownership, not IT ownership. McKinsey reports that high performers are about three times more likely to show strong senior-leadership ownership and engagement in AI. BCG's playbook opens with "lead from the top" — a multi-year strategic AI ambition owned by the CEO, not delegated to a technology function. Accenture's Reinventors are led by executives who treat technology change as a business strategy rather than an IT project. The pattern holds: when AI reports to IT, it optimises technology; when it reports to business leadership, it optimises outcomes.
Governance that scales, not governance theatre. McKinsey finds that AI incidents are common — around half of organisations report at least one — and that high performers manage the risk with human-in-the-loop rules, centralised oversight, and executive accountability rather than a policy PDF. Deloitte's finding that only about one in five firms has a mature model for governing autonomous agents tells the other side of the same story: governance has not scaled with deployment. The trust barrier analysis examines why governance is the operational bottleneck, not a compliance formality.
Capability building, not tool purchasing. Every study distinguishes between buying AI tools and building AI capability. The distinction is not semantic. Tools are software individual employees use. Capabilities are organisational muscle — the data infrastructure that makes enterprise knowledge accessible to AI systems, the evaluation frameworks that measure AI against business outcomes rather than demo quality, the feedback loops that improve systems from production data, and the people who can design and run AI workflows rather than merely operate AI interfaces.
Why the gap is widening
The compounding effect explains the divergence. BCG's 3.6-times total-shareholder-return differential is not a one-year accident. It compounds: each redesigned workflow generates data that improves the next workflow, builds muscle that accelerates the next deployment, and produces returns that fund the next investment. The minority that crossed the structural threshold are accelerating. The majority are iterating on pilots — and Bain's data is explicit that this is where most enterprises remain stuck, with gains plateauing rather than scaling.
The talent constraint amplifies the divide. Deloitte names the skills gap as the single biggest barrier to integration. That is a structural problem, not a temporary one. Organisations that transformed early attracted and developed the people who enable continued transformation. Those that delayed now compete for AI engineering, data architecture, and AI product-management skills that are scarce and expensive. Talent follows impact, and impact is concentrating where it already exists.
The governance burden scales non-linearly. As organisations deploy more AI systems, governance requirements multiply rather than add. Each system needs monitoring, each agentic workflow needs delegation rules, each model update needs validation. Firms that built governance into their first deployments have a scalable foundation; firms retrofitting it after the fact face complexity that grows faster than the deployment that caused it. The cost structure analysis shows how governance costs compound when treated as an afterthought.
What the synthesis means for DACH enterprises
The investment-returns gap is not an American phenomenon observed from a distance. DACH enterprises face the same dynamics with two additional realities. The first is regulatory: from 2 August 2026, the bulk of the EU AI Act applies, including the obligations for high-risk systems in Annex III — conformity assessment, technical documentation, logging, human oversight, and registration. For a Mittelstand firm, governance that scales is not optional polish; it is the difference between a deployable system and a compliance liability. The second is cultural: the Mittelstand's traditional strength — operational excellence within well-honed established processes — quietly becomes a liability when the lesson from all five studies is that those very processes have to be redesigned, not optimised.
The path from the majority to the minority is not more AI. It is a different operating model for AI. The five studies converge on the same requirements: workflows redesigned around AI capabilities rather than augmented with AI tools; business leaders who own AI outcomes rather than technology leaders who own AI infrastructure; governance architecture that enables scaling — and, for DACH firms, satisfies the AI Act — rather than policy documents that merely constrain; and capability building that creates organisational muscle rather than procurement that creates software licences.
The question is not whether to invest in AI. That decision is made — 88 to 95% of enterprises already are. The real question is whether the investment is structural or incremental. Structural investment changes how work flows, how decisions are made, and how the organisation learns from AI-generated data. Incremental investment adds AI to existing processes and hopes the productivity gains aggregate into financial impact. Five independent studies, across tens of thousands of organisations, agree that they do not.
Every article in this series examines one dimension of the gap — the McKinsey scaling analysis, the BCG maturity framework, the Accenture reinvention premium, the trust infrastructure required to scale, and the cost structures that decide whether investment compounds or dissipates. This synthesis makes the unified argument: the gap is real, it is widening, and closing it requires not more AI spend but a fundamentally different way of deploying it.
A Fit Call tells you, in 30 minutes, whether your current AI investment is structural or incremental — and which specific gap (workflow design, ownership model, governance architecture, or capability foundation) is keeping your spend from compounding — before another budget cycle plateaus.
References: McKinsey & Company, "The State of AI: How Organizations Are Rewiring to Capture Value," 2025; McKinsey Global Institute, "The Economic Potential of Generative AI"; BCG, "The Widening AI Value Gap," September 2025; Accenture, "Reinvention in the Age of Generative AI"; Deloitte, "State of Generative AI in the Enterprise"; Bain & Company, "Survey: Generative AI's Uptake Is Unprecedented Despite Roadblocks"; European Commission, "AI Act implementation timeline".