Why leadership's own usage decides whether the license creates value. Audience: board members and CEOs.
> 📥 The full article is available as a PDF with models, sources, and appendices. Download it at the top of the page.
The license is paid. The question isn't technological.
Many large Danish companies opened up GenAI tools — Microsoft Copilot, ChatGPT Enterprise, sector-specific platforms — in 2024 and 2025. Denmark leads the EU on enterprise AI use: Eurostat shows Denmark had the EU's highest share of AI-using companies in 2025, and the Confederation of Danish Industry's 2026 report states that 75% of Danish companies with 250+ employees used AI in 2025. Budgets are approved. Governance policies are written. IT has set the security perimeter. On paper, the company is in motion.
But one question rarely surfaces in the boardroom: Does the CEO actually open the tool in the morning?
It isn't ironic. It's the single variable McKinsey and MIT have identified as one of the clearest differentiators of whether the AI investment actually moves EBIT.
The numbers a board should worry about
McKinsey's State of AI 2025, published November 2025, shows 88% of organizations use AI in at least one function. Only about 6% qualify as AI high performers, where AI contributes more than 5% of EBIT. The difference between the 6% and the other 82% of AI users is not the technology stack. It's execution and ownership.
MIT's NANDA report from summer 2025 tells the same story from another angle. 95% of enterprise GenAI pilots deliver no measurable P&L impact. The cause lies in integration, workflow redesign, and organizational ownership. Model quality is rarely the problem.
McKinsey is precise about what moves the numbers. AI high performers are 3x more likely to strongly agree that their senior leadership demonstrates ownership and commitment to AI initiatives. That means active, visible use. Role-modelling that the CEO actually opens the tool.
Dataiku's Global AI Confessions Report from March 2025 documents the other side. 94% of CEOs suspect their employees are already using GenAI tools without approval. That's shadow use. And it's a governance risk that arises precisely when leadership doesn't take ownership of the agenda itself.
Why leadership use is decisive
Organizations where the CEO and executive team use AI in their own daily work create an adoption cascade. Middle managers see it. Employees see it. "Have you asked Copilot first?" becomes a natural part of meeting culture. Usage data climbs. Cycle times drop. Results become measurable.
Organizations where leadership doesn't use the tool get the opposite dynamic. A few enthusiasts experiment. Everyone else doesn't. Or they do — in secret — which is the worst possible governance situation.
A Danish case: Royal Unibrew
Royal Unibrew is not an AI-native startup. It's Denmark's second-largest brewing group, with roots back to 1856, listed on Nasdaq Copenhagen, with brands like Faxe Kondi, Royal Beer, Ceres, Lapin Kulta, and Original Long Drink. In 2025 they posted net revenue of DKK 15.7 bn (+5%) and EBIT of DKK 2.2 bn (+12%). EBIT margin expanded 90 basis points to 14.0%. ROIC rose to 13%.
This isn't a risk-free growth story. In April 2026, PepsiCo announced certain agreements with Royal Unibrew would not continue beyond 2028, and the stock was hit hard. The point of this article is not that AI alone explains financial performance, but that the case shows a large Danish company that has translated AI from inspiration to concrete operating model.
In January 2024, the group's Danish marketing department held an internal AI inspiration day. Over the following months, in collaboration with the Danish firm Manifold AI, they developed five AI agents. Each has a name, a face, a specific role, and an email address. All are integrated in Microsoft Teams. They are not experiments. They are in daily operation.
Royal Unibrew's five AI agents
Kondi Kai — brand agent. Named after the iconic Faxe Kondi. Fed 50 years of tone-of-voice and brand guidelines. Acts as an internal sparring partner for the brand manager and gives feedback on agency proposals.
Athena — market analyst. Most-used of the five. Analyzes market data and reports, giving marketing and insight teams faster access to analyses. Michala Svane, Director of Marketing, Digitalization & Business Development: "Our go-to-market is much faster, because we get faster access to data and can do complex analyses much, much faster."
Prometheus — sales forecaster. Statistical modelling across internal sales data, external databases, and seasonal variation. Forecasting, inventory optimization, quarterly forecasts.
Møller — sommelier. Food-beer pairing. Staff training, brand consistency across markets, hospitality-oriented customer experience.
Ella — trade specialist. Hospitality sector. Pricing optimization, on-trade insights, channel performance.
> When the agents got names and faces, internal use and engagement rose 4x.
The interesting lesson for boards is about behavior, not technology. Humanizing the agent was the actual adoption lifter.
Two international references
Satya Nadella, Microsoft. In August 2025, Nadella published his five standing GPT prompts. One: "Based on my prior interactions with [person], give me five topics likely to be top of mind for our next meeting." It's a productivity habit, not strategic signalling.
Stéphane Bancel, Moderna. Bancel expects employees to use ChatGPT Enterprise at least 20 times per day. Within two months of rollout, Moderna had over 750 custom GPTs across the company. 83% of employees used ChatGPT daily. Bancel's framing: "If we had to do this the old biopharmaceutical way, we would need a hundred thousand people today."
The three levels of executive AI use
Three distinct levels. They build on each other. Don't jump to level 3 before level 1 is a daily habit.
Level 1 — Personal productivity. Email summarization. Draft replies. Briefing before meetings. Transcription and action points. It requires one thing: opening the tool in the morning.
Level 2 — Leadership preparation. Upload board decks. Have AI play three personas: aggressive growth, conservative risk, and the most critical owner. Red-team before owner meetings. Headwind before the meeting, not after.
Level 3 — Strategic decision-making. Scenario analysis for M&A and market entry. Wargaming against competitor personas. Experiments with multi-agent simulations as decision support.
Five use cases that are typically underestimated
1. Decision preparation with pushback. Upload material. Ask for three counter-arguments, the worst realistic scenario, and the three questions a sceptical CFO would ask.
2. Board decks with persona stress-test. Run draft decks through several LLMs playing different board members.
3. Red-teaming before CEO and owner meetings. Run the deck through AI with the counterparty as a persona. PwC's 2025 survey: only 2% of directors cite pressure-testing as actual practice.
4. Competitor wargaming. Three top competitors, three moves they could make next quarter, based on public statements and earnings calls.
5. Translating strategy into three variants. Middle manager, employee, investor. If the strategy can't be explained simply, it won't be executed either.
The board's seven questions
Seven questions every board should be able to answer clearly by the end of the next meeting. They are not technical. They are governance questions.
1. Does the CEO and executive team actually use the tool daily? Not as a budget line. As a personal habit, observable in their own workflow.
2. Are there measurable business cases, not just deployment metrics? Number of licenses is input. EBIT impact, time saved, cycle time is output.
3. Who in the executive team owns the AI agenda? If it's "everyone", it's no one. One executive must be held accountable with KPIs.
4. What is the governance frame for AI in strategic decisions? Where can agents make decisions? Where do we require human review?
5. What is our exposure if we don't accelerate? Competitors 12 months ahead win on cycle time. Quantify the risk.
6. How do we measure whether adoption is real or cosmetic? License utilization, daily active users, prompts per employee. Shadow use is a red flag.
7. Do we have the capacity to oversee this ourselves? KPMG/INSEAD's 2026 AI Board Governance Principles: nearly three quarters of boards are assessed to have only moderate or limited AI expertise.
The license is paid
The difference between winners and losers in AI transformation is not just budget, model choice, or governance policy on paper. It lies in whether leadership translates AI into redesigned workflows, measurable business cases, and visible daily use.
We are past the hype phase. We are in the execution phase. Boards that don't put the seven questions explicitly to their CEO and executive team risk missing one of the most important execution disciplines in the coming years of digitalization.
The question is no longer whether AI works, but whether your company has actually put it to use.
📥 Download the full article as a PDF with models, diagrams, and complete source list at the top of the page.
🎓 Want a structured conversation in your board or executive team? Book an AI Readiness Sprint or an executive workshop.
📬 Subscribe to the newsletter "AI, Built Human" on Substack — weekly insights on AI in practice.
Stefano Vincenti · AI Advisor & Trainer · aitrainer.dk · External Lecturer, IT University of Copenhagen · Cofounder & CTO BotTellMe · Partner, TryZone