The missing agenda item: why AI governance does not reach the board’s agenda

From 24 interviews with board chairs and directors: a pattern that runs across every sector, and three questions that help to address it.

Over the past months I have spoken with 24 board chairs and directors so far, women and men with more than 75 active mandates between them, across 15 sectors, from the large bank to the family firm. One question ran through almost every conversation: how does AI actually reach the board’s agenda? The most frequent answer, in varying words: not really at all. The topic is present, but it is not anchored.

This is the pattern I call the missing agenda item.

The missing agenda item

AI surfaces in many board meetings, but rarely as an agenda item in its own right. It is touched on when management reports on efficiency projects. It is mentioned in the risk report, but rarely broken down into specific categories. Yet as an agenda item, with a decision point, a board paper, a resolution, minutes? That is rather rare.

One of the directors I interviewed, who sits on the boards of several listed companies, described it like this: AI comes up at every meeting, but it has not yet truly calibrated itself, it is still somewhat in flux. Another director, with mandates on Swiss and German bodies, said unprompted that boards somehow cannot get their foot in the door.

What is striking is this: these board members do not reject the topic. They are interested, many of them personally enthusiastic. But between interest and agenda item there opens a gap, and it is precisely this gap that costs governance substance.

A board chair explained the mechanics behind this to me. On a board, he said, a topic only becomes truly real once it lies on the table as a board paper. There is no informal space in between in which something can quietly mature. But the moment a topic is placed on the agenda, it becomes a priority, because everything on a board’s agenda is a priority. And so a prudent chair faces a choice: place AI formally on the agenda, and risk lifting it above other matters before its time; or leave it off, and let it remain invisible. Many resolve this by deliberately keeping the topic small until it is ripe. That is understandable. And it is precisely the mechanism that produces the missing agenda item.

For what is not placed on the agenda is not decided. What is not decided runs in day-to-day business. And what runs in day-to-day business is, by definition, no longer a board matter but a management matter. The consequence: decisions relevant to AI are taken without the board ever having addressed them explicitly. About vendor lock-in. About whether board papers disclose when they were produced with AI support. About which AI risks are even monitored in specific categories. And ultimately about the company’s AI strategy, which forms out of day-to-day business instead of being led by the board.

Not knowledge, but questions

Why does the topic fail to land? The obvious answer would be: because board members understand too little about AI. This explanation features prominently in the literature, and it is not wrong. But in my view it is incomplete.

What the interviews show is more subtle. Anyone who does not know the right questions cannot judge whether what management reports is sufficient. An experienced director, asked directly whether his board was well enough equipped to put the critical questions, told me: I would say no, we are not there yet, we are on our way. Another, with experience in building governance frameworks, put it this way: boards want to ask the right questions, but he did not know whether they knew what the right questions were.

This is not the same as a lack of AI competence. A board does not need to understand how a large language model is trained in order to decide whether it should be deployed. It needs to understand which question to put to management in order to derive a judgement from the answer. These are two very different competences, with very different costs of acquisition.

As long as the repertoire of questions is missing, what should not happen does happen: the board delegates, often without noticing. Not out of convenience, but because it simply does not know where it could intervene.

Here lies a tension I encountered several times. AI can genuinely strengthen the board: those who use it themselves arrive at sharper questions, place board papers in better context, test assumptions faster. Several interviewees confirmed this to me. But this benefit has a precondition. To use AI meaningfully takes a minimum of personal familiarity with the tool: enough to ask usefully, and enough to tell a good answer from one that is merely plausible.

It is precisely this minimum that many lack. An experienced director from the financial sector said it plainly: across boards in Switzerland, the competences for dealing with technology risk are on average not strongly developed, and there is a relatively large group that has never properly thought the topic through and simply goes along with it.

AI is a lever, but you need a footing from which to move it. And here lies the uncomfortable truth from my conversations: for the majority of boards, the information asymmetry towards their own management is growing, not shrinking. Not because these boards are incapable, but because the equipment is missing and no one is putting it in their hands. The tool meant to close the gap widens it, as long as the equipment is missing.

What would help on Monday, at the next meeting

If my observation is right, that the missing agenda item is not a problem of AI knowledge but a problem of questions, then three concrete questions are a good start.

Question one: where is our management already using AI operationally, and who decides on its deployment? This question is not trivial. The interviews showed me repeatedly that even the boards of large companies had no complete picture of where, in their own house, AI is already in use, let alone who decides in each case. Taking stock is the precondition for any governance discussion.

Question two: how is our organisation set up around AI, centralised, decentralised or hybrid, and where does the decision boundary run between the board and management? This question was described repeatedly in the interviews as central, by some even as the actual core of governance work on AI. It forces the delegation boundary to be made explicit, instead of leaving it implicit. And an implicit boundary is exactly what makes the missing agenda item possible in the first place.

Question three: what does management not bring us on AI, which opportunities, which projects, which risks do we not see at all, because they never appear on a board paper? A board chair described it to me like this: what he, as a director, does not see is what is never made or never put forward in the first place. The board can only examine what lies before it. This question is meant to direct attention to what the other two do not reach.

These three questions are no substitute for a question framework. But they open the conversation, and they ensure that next time AI does not slip through the net but reaches the agenda. With a decision point, a board paper, a resolution, minutes.

Exactly where it belongs.

This article first appeared on the blog of SwissBoardForum, the leading platform for board directors in Switzerland.