Question 1

How should a board of directors govern artificial intelligence?

Accepted Answer

By treating AI as a delegation of power rather than a tool: mapping the agents deployed, identifying technological dependencies, setting a framework compliant with the AI Act, and entrusting ongoing oversight to a dedicated body of the board.

Question 2

What questions should a board ask its executive management about artificial intelligence?

Accepted Answer

Four are enough to start. Where is AI shifting value in our sector, and which side are we on? Which agents are already deployed, by whom, and on which decisions? Which suppliers do we depend on, and what would it cost to switch away? What measurable objective serves each use case, and what has it delivered? A board does not need to understand the technology. It needs answers that hold up.

Question 3

Who bears the risk linked to artificial intelligence within the company, and does the board know it?

Accepted Answer

Most often it is spread across IT, legal, security, business lines and suppliers, with no one accountable to the board. The first question to ask is therefore one of naming: who, specifically, reports on AI-related incidents, and above what threshold is the board alerted?

Question 4

Our board has no expertise in artificial intelligence: should we recruit a director or train our own?

Accepted Answer

The two address different needs. Training gives the board the vocabulary and the right questions, in a few sessions. Recruitment brings permanent expertise and an outside perspective on trade-offs. The criterion: if AI affects the business model and not just processes, that expertise needs a seat at the table.

Question 5

How do you put artificial intelligence on a board's agenda without turning it into a technical topic?

Accepted Answer

By framing it where it has its effects: strategy, risk, dependency and accountability. An agenda item about models and tools loses the board. An agenda item about value shifting, decisions handed to a machine, suppliers you can no longer leave, and liability in the event of an incident keeps their attention.

Question 6

What should an artificial intelligence dashboard presented to the board contain?

Accepted Answer

Five lines are enough. A map of the systems deployed and the decisions they make. Supplier dependencies and the cost of switching away. The measurable objective of each use case and its actual result. The number of errors detected and corrected by a human. The state of compliance with the AI Act.

Question 7

Our AI committee is both judge and party, how do we get out of that?

Accepted Answer

A committee that chooses use cases, executes them, and then judges its own success does not produce an arbitration, it produces its own justification. But the real issue lies upstream: this committee exists because AI was treated as a technical matter to delegate. Thinking through what AI means for the company is neither an IT matter nor a human-resources matter. It is a work of strategic architecture that belongs to the executive. Until the arbitration is escalated there, the committee will remain judge and party, whatever its composition.

Question 8

Our artificial intelligence projects stay stuck at the pilot stage and never scale, what should we do?

Accepted Answer

The issue is not the pilot. A test succeeds in isolation because it is dropped onto an organisation designed before AI, without changing anything in it. Yet every task taken over by a machine shifts a boundary: who decides, who checks, who is accountable. As long as those boundaries do not move, the gain dissipates at the seams and nothing shows up in the numbers. What needs to be examined before talking about scaling: what work disappears, what work appears, who takes over the decision the machine prepares, and what stops being done. This is a work of architecture. It belongs to the executive, and it cannot be delegated to either IT or HR.

Question 9

Our departments are not moving at the same pace on artificial intelligence, who should own the topic?

Accepted Answer

The executive. A gap in pace between departments signals that the architectural work has not been done, and that each was asked to move forward without knowing towards which organisation. Delegating the topic to any one department amounts to letting the slowest one set the pace for all the others.

Question 10

How do you link each artificial intelligence use case to a measurable objective?

Accepted Answer

By asking three questions before any deployment. Which decision or which process does this use case change? Which indicator will move, in which direction, by how much? Who stops it if the effect is negative, and at what threshold? The most common trap lies elsewhere: many companies have deployed AI without seeing their margins move, and the bottleneck rarely lay in the tool. It lay in where it was plugged in, within an organisation designed before AI. A time saving that does not change the process around it never shows up in the margins.

Out of the fog, understanding AI.

Where to start?

Defend, attack, or both

What AI can do right now

The 4 D rule

Behind every use case, decisions

White paper

La règle des 4 D

The four generations of artificial intelligence

AI in the enterprise: the architectural work that falls to the leader

One topic, one hour, one conversation.