In short: An advisory or supervisory board doesn’t need to understand AI technically to oversee it effectively. It needs to ask five questions that pull AI out of IT and make it a matter for top management: What is AI changing about our business model? Does our data hold up? What does AI cost per unit of work done? Who decides what a system may do on its own? And is our organization built for what we’re planning?
Why is AI a topic for advisory and supervisory boards?
Because AI now touches business models, cost structures and liability, which are exactly the areas a board exists to oversee. Boards are catching up, but slowly.
66%
In the second edition of Deloitte’s “Governance of AI” study, with 695 respondents from 56 countries, 66 percent say their board has limited or no knowledge of and experience with AI.
For 31 percent, AI isn’t on the agenda at all, down from 45 percent in the previous survey (Deloitte Global Boardroom Program). In the US, according to NACD, more than 62 percent of directors at public companies surveyed say their board now gives AI dedicated time on the full board’s agenda (NACD 2025 Board Practices and Oversight Survey).
Mid-sized companies have an additional wrinkle. In a German GmbH, an advisory board has no statutory duties. Its role is defined by the articles of association and its rules of procedure. It can be purely advisory or hold approval and oversight rights. Either way, an advisory board that doesn’t raise AI is missing the point where the company’s next few years will be decided.
Question 1: What is AI changing about our business model, not just our processes?
Most AI presentations are about efficiency: faster quotes or automated reports. That’s true, but it’s the smaller half. The bigger question is whether AI changes what customers pay for, who competes with the company and which assumptions behind its success still hold. A software company that charges per user has a different problem from an equipment maker whose service work could be done by agents.
How the board can tell a good answer: Management can name which share of revenue depends on assumptions that AI calls into question, and which competitors or customer-built solutions it’s watching.
Question 2: Does our data support what we plan to do with AI?
AI is only as good as the data it can access. In many companies, data has grown over decades, scattered across systems, incompletely maintained and only partly digital. An agent that is supposed to act reliably needs current, structured data it is also allowed to use. Skip that step and you end up funding pilots that work in the demo and fail in production.
How the board can tell a good answer: There is an honest assessment of data quality for the two or three most important use cases, with an owner and a plan. A data strategy on slides isn’t enough.
Question 3: What does AI cost us per unit of work done, and who measures it?
AI brings variable costs back. Every request to a language model costs money, and the costs vary widely with the difficulty of the case. Gartner points out that additional AI features often increase costs without improving results (Gartner, July 1, 2026). A board should therefore ask about the cost per completed case and the benefit on the other side of the ledger. The AI budget alone says little.
How the board can tell a good answer: Costs are known per use case, there are caps, and impact is measured against a metric that was captured before the project started.
Question 4: Who decides what an AI system may do on its own?
As soon as AI acts on its own, placing orders, scheduling service calls or answering customers, autonomy becomes a leadership decision. It should never quietly expand just because a pilot went well. It needs an explicit rule: where does the system act on its own, where does it only suggest, who approves, how is it logged and how often is this reviewed?
How the board can tell a good answer: There is a written rule for the autonomy of the systems in use, a person accountable for it, and logs that make it possible, if in doubt, to trace what a system did and why.
Question 5: Is our organization built for what we’re planning?
The most common brake on AI lies in the organization. Incentives reward the old business, roles are cut out for routine work, and leadership knows AI from presentations but not from using it. A board can make more of a difference here than on any technology question, because it can raise with management the issues nobody raises internally.
How the board can tell a good answer: Management uses AI tools itself, compensation for key roles fits the new direction, and there is a plan for the employees whose work is changing the most.
How does a board anchor these questions for the long term?
A standing agenda itemAI once a quarter as its own item, not as a sub-item of the IT report.
A few metricsfor example the share of revenue tied to assumptions at risk from AI, the cost per completed case and the rate at which employees accept AI suggestions.
In-house expertiseAt least one board member should know AI from operational responsibility. Everyone else should have the most important applications demonstrated to them at least once.
Frequently asked questions
Does an advisory board need AI expertise?
An advisory board doesn’t need technical command of AI. It needs enough understanding to assess opportunities, costs and risks and to evaluate management’s answers. Having at least one member with hands-on AI experience significantly raises the quality of oversight.
What’s the difference between an advisory board and a supervisory board on AI?
The supervisory board of a German stock corporation (AG) has a statutory duty to oversee management. An advisory board in a GmbH only has the rights and duties granted by the articles of association or its rules of procedure. On AI, the questions are similar. How binding they are differs.
How often should an advisory board discuss AI?
At least once a quarter as a separate agenda item. More often in phases when AI is changing the business model, plus an annual in-depth session on strategy, investments and risks.
This article does not replace legal advice on the duties of advisory and supervisory board members.
Further reading: Evaluating AI investments · EU AI Act: what advisory and supervisory boards need to know · Governance starts with trust
Sources: Deloitte (2025): Governance of AI, 2nd edition · NACD (2025): Board Practices and Oversight Survey, AI · Gartner press release, July 1, 2026