Why AI value is a leadership question — the ownership gap behind the adoption paradox.
McKinsey’s latest global survey finds 88% of organisations using AI and about 6% able to attribute meaningful profit to it. The practices that separate the two groups are managerial — and Switzerland, which leads Europe on daily use at work, sits on the same side of the divide.
Les Insights sont publiés en anglais.
McKinsey published the latest edition of its Global Survey on the state of AI on 5 November 2025. One figure from it has circulated widely since: 88 per cent of respondents say their organisation regularly uses AI in at least one business function, against 78 per cent a year earlier. The figure printed beneath it has travelled much less. Thirty-nine per cent attribute any level of EBIT impact to that use, and most of those say the share concerned is under 5 per cent.
The survey then isolates a smaller group: respondents who attribute 5 per cent or more of EBIT to AI and who also report significant value from it. McKinsey calls them AI high performers. They are about 6 per cent of the sample.
What separates them is set out in the same tables. Almost none of it is a purchasing decision.
Six per cent is a measurement threshold, not a verdict on technical skill
The survey ran from 25 June to 29 July 2025 and collected 1,993 responses from 105 nations, weighted by each nation’s contribution to global GDP. Thirty-eight per cent of respondents work for organisations above one billion dollars in annual revenue: a large-company picture, and the constraint it identifies is not one a bigger budget dissolves.
Most respondents are not yet where enterprise-level effects could appear: nearly two-thirds say their organisation has not begun scaling AI across the enterprise, and roughly one-third has. Scale tracks size: nearly half of respondents at companies above five billion dollars in revenue have reached the scaling phase, against 29 per cent of those below one hundred million.
Value is being reported, simply not in the accounts: what is missing is the step from a use case that works to a number a finance director will sign.
What separates the six per cent is a short list of management decisions
The survey tested which practices distinguish them, and the answers are concrete.
They are three times more likely than their peers to strongly agree that senior leaders demonstrate ownership of and commitment to AI initiatives, and much more likely to say those leaders drive adoption personally, including by using the tools themselves. They are more than three times more likely to say their organisation intends to use AI to bring about transformative change, and nearly three times as likely to have fundamentally redesigned individual workflows — the practice McKinsey singles out as having one of the strongest contributions to business impact of all the factors tested.
Two others sit alongside it. High performers more often have defined processes for when model outputs need human validation — another top distinguishing factor — and they resource the work: more than a third commit over 20 per cent of their digital budget to AI, and about three-quarters are scaling or have scaled, against one-third of everyone else. Eighty per cent of all respondents set efficiency as an objective; high performers more often add growth or innovation.
Two cautions belong with these numbers: they are self-reported perceptions, and the survey reports associations, not causal effects. The validation practice also has a defensive value: 51 per cent of respondents at AI-using organisations report at least one instance of a negative consequence, and nearly a third of all respondents report consequences stemming from inaccuracy.
The direction matches McKinsey’s wider organisational research. The State of Organizations 2026, published on 19 February 2026 from more than 10,000 senior executives across 15 countries and 16 industries, reports technology leading organisations to “reimagine how work gets done, redefine domains and end-to-end processes, and rethink traditional structures”. Writing in Harvard Business Review on 30 April 2026, partners at Bain & Company and economists at OpenAI recorded the same gap between widespread adoption and commensurate bottom-line improvement.
Switzerland leads Europe in daily AI use and sits in the same single digits for depth
The Swiss position sharpens the point, because it removes the usual explanation.
Accenture’s Pulse of Change survey, conducted between April and June 2026 among 980 senior executives and 980 non-management employees in Europe across 19 sectors — 100 of each in Switzerland, a national sub-sample of 200 — found 49 per cent of Swiss respondents using AI tools daily against a European average of 28.3 per cent. Eighty-eight per cent report full or moderate access to employer-provided AI tools, against 80 per cent globally.
Organisational depth is another matter. EY Switzerland’s survey of 604 respondents, published on 27 May 2026 and relayed by the Confederation’s SME portal, found 89 per cent using AI in their working day. But only 32 per cent say their company deploys it in individual business areas, 14 per cent that it scales systematically across several, and 9 per cent that AI has been integrated into the core of the business in a way that changed the model. Thirty-one per cent are still in pilots or proofs of concept; 14 per cent have started nothing concrete. Twenty-nine per cent may use AI through a private account, and 8 per cent have nothing else.
That sample tilts towards large employers, 30 per cent of them above 10,000 employees, which makes the share still in pilots harder to attribute to a shortage of resources. The two surveys do not measure the same thing — EY asks about business-model change, McKinsey about attributable profit — but both land in single digits for the firms that have genuinely reorganised around the technology.
European law has begun to describe ownership in the same terms
Obligations under the Artificial Intelligence Act attach to the deployer — the organisation using an AI system — and not only to the vendor. Article 26(2) of Regulation (EU) 2024/1689 provides that deployers of high-risk systems “shall assign human oversight to natural persons who have the necessary competence, training and authority, as well as the necessary support”. Those high-risk obligations now begin on 2 December 2027 for the Annex III use cases and on 2 August 2028 for AI embedded in regulated products, after the Digital Omnibus on AI, Regulation (EU) 2026/1744.
Article 4, as replaced by that regulation, is broader: providers and deployers must take measures to support the development of AI literacy among staff and others operating AI on their behalf, without having to guarantee any level for any individual. Its recital states that AI literacy should be a strategic priority regardless of regulatory obligations. Swiss firms are reached through Article 2(1)(c), which extends the Act to third-country deployers where the output produced by the AI system is used in the Union.
The decisions that do not require a new budget
Nothing in these findings points to better technology in the companies reporting value. The distinguishing practices are available to a firm of twelve people.
- Redesign one workflow end to end rather than adding a tool per department. This is the practice most strongly associated with measurable impact.
- Give it an owner with authority over how the work is done, not only over the licence budget — the ownership the survey measures and the oversight the AI Act will require of high-risk deployers.
- Write the validation rule before scaling: which outputs need a human check, by whom, and before what leaves the company. Half of respondents at AI-using organisations already report one negative consequence or more.
- Settle the private-account question. In the Swiss sample, 29 per cent may use AI through a private account, 8 per cent with nothing else. Either can be a policy; neither should be an accident.
- Add one objective beyond efficiency. Efficiency is near-universal as a goal; growth and innovation are what the higher-performing group adds to it.
- Fix one number before and after — quotation turnaround, first-response time, order-entry error rate. Enterprise EBIT is where value appears last, not first.
The dates that will carry information
The regulatory calendar is already set. Article 4’s literacy obligation has applied since 2 February 2025; the high-risk obligations, including the assignment of human oversight, arrive on 2 December 2027 and 2 August 2028. The annual editions of these surveys — McKinsey’s in the autumn, the Swiss readings in spring and summer — are where movement in the single-digit share would first show.
Until then, the useful question inside a company is narrower than the technology debate around it: who here owns the way this work is done, and have they decided anything yet?