McKinsey's 2025 State of AI survey delivers the clearest empirical picture we have of enterprise AI adoption — and the results should alarm every executive who believes adoption equals impact. The survey covered 1,993 participants across 105 nations. The headline finding: 88% of organisations now use AI in at least one business function, and 72% use generative AI specifically, up from 33% in 2024. Adoption has reached near-universality.

But here is the number that matters: of those 1,993 respondents, only 109 — exactly 6% — attribute more than 5% of EBIT impact to their AI initiatives. McKinsey classifies these as "high performers." The remaining 94% are using AI without material financial impact.

This is not a technology problem. The tools are available to everyone. The models are commodity. The APIs are accessible. The 6% and the 94% have access to the same technology stack. What separates them is how they deploy it — and the data on that distinction is unambiguous.

The workflow redesign gap

The single strongest predictor of AI-driven EBIT impact in McKinsey's data is workflow redesign. High performers are 2.8 times more likely to have fundamentally redesigned workflows around AI capabilities — 55% of high performers versus 20% of the rest.

Nearly 80% of organisations layer AI on top of existing processes without rethinking how the work flows. They take the current process, add an AI step, and expect transformation. What they get is incremental improvement — a faster version of a process that was designed for humans, not for human-AI collaboration.

The distinction is structural, not cosmetic. Layering AI onto an existing claims processing workflow means the AI drafts what a human would have written. Redesigning the workflow means the AI classifies, routes, and drafts simultaneously, while humans review exceptions and handle complex cases. The first approach saves 15 minutes per claim. The second approach changes the throughput of the entire operation.

This maps directly to the Three Levels framework. Layering AI onto existing processes is Level 01 — AI as tool. Redesigning workflows is Level 02 — AI as specialist. McKinsey's data confirms what the framework predicts: Level 01 produces measurable but non-material gains. Level 02 produces EBIT impact.

Leadership is the second differentiator

McKinsey's high performers share a second characteristic: their leaders are three times more likely to demonstrate visible ownership and commitment to AI initiatives. This is not about giving speeches about digital transformation. It is about executive sponsors who own specific workflow outcomes, allocate dedicated budget, and review operational metrics weekly.

The leadership gap explains why most AI programmes stall after the pilot phase. Pilots succeed because they have a champion — someone who pushes through obstacles, secures data access, and protects the team's time. Scaling from pilot to production requires that same intensity applied to organisational change: redefining roles, reallocating capacity, changing how teams are measured. Without executive ownership, the organisation defaults to its existing operating model, and the AI initiative reverts to tool usage.

For DACH Mittelstand companies, this pattern is particularly relevant. The Geschäftsführer's involvement is not optional — it is the mechanism that makes workflow redesign possible. In a 400-person industrial supplier, the distance between the executive and the operational team is small enough that direct ownership can drive change in weeks rather than quarters. This structural advantage is available to every mid-market company. Most do not use it.

Where the value concentrates

McKinsey's data reveals where AI creates measurable impact and where it remains aspirational. Software engineering and IT functions report 10 to 20% cost reductions — the most mature deployment category, where code generation, test automation, and infrastructure management have well-defined workflows. Marketing and product development show 10% or more revenue uplift, driven by personalisation, content generation, and accelerated development cycles.

The pattern is consistent: functions with structured, repeatable workflows capture value. Functions where AI is used ad hoc — for brainstorming, research, or general productivity — show gains that are real to individuals but invisible on the income statement.

McKinsey's baseline estimate of $2.6 to $4.4 trillion in annual value potential across 63 use cases remains directionally valid. But the survey data shows that potential is being captured unevenly. The 6% are concentrating returns in high-structure workflows. The 94% are spreading AI thinly across low-structure tasks.

The risk landscape compounds the problem

Adoption without redesign creates a specific risk profile. McKinsey reports that 74% of respondents cite inaccuracy as a top risk and 72% cite cybersecurity concerns. These risks are not inherent to AI technology — they are symptoms of how AI is deployed.

Inaccuracy risk is highest in unstructured deployments. When AI is used as a general-purpose tool without defined inputs, expected outputs, or validation steps, errors propagate undetected. When AI is embedded in a redesigned workflow with defined delegation rules, confidence thresholds, and review cycles, inaccuracy is caught and corrected systematically. The delegation and review framework exists precisely to manage this risk operationally rather than hoping it does not materialise.

Cybersecurity risk follows a similar pattern. Unmanaged AI usage — employees using consumer AI tools with company data — is inherently harder to secure than governed AI workflows running through defined data pipelines with access controls.

The companies in the 6% do not experience fewer risks. They manage risk structurally, through workflow design, rather than reactively, through policy documents that employees ignore.

What this means for DACH Mittelstand

McKinsey's data confirms a pattern we observe across the DACH mid-market: the gap between AI adoption and AI impact is not closing. It is widening. More tools, more licences, more training — none of this moves the needle unless the underlying workflows change.

The diagnostic question is not "Which AI tool should we buy?" It is "Which workflows should we redesign first?" The answer depends on three factors: the volume and structure of the workflow, the accessibility of the required data, and the presence of an executive sponsor who will own the outcome. These are the readiness dimensions that determine whether an AI initiative produces EBIT impact or another pilot that quietly fades.

The AI Operating System methodology is built on the same principle that McKinsey's data validates: workflow redesign, not technology selection, determines whether AI creates operating leverage. The six-dimension diagnostic evaluates exactly the factors that separate the 6% from the rest — workflow readiness, data accessibility, decision authority, and operating model clarity.

Most DACH mid-market companies sit in what McKinsey's data describes as the 94%: using AI, measuring adoption, reporting progress, but seeing no material impact on the income statement. The transition to the 6% does not require more technology. It requires someone to decide which workflow changes first — and to own that decision.

Use the diagnostic to identify your highest-leverage workflow for AI redesign. We assess readiness across all six operational dimensions and identify whether you are building adoption or building impact — and what it takes to move from one to the other. Start your diagnostic →

References: McKinsey & Company, "The State of AI: How Organizations Are Rewiring to Capture Value," Global Survey, November 2025 (1,993 participants, 105 nations); McKinsey Global Institute, "The Economic Potential of Generative AI," June 2023 ($2.6–4.4T value estimate across 63 use cases).