AI: a governance question before it is a technology question
Surveys of directors converge: AI is the topic on which boards feel least equipped. What a holding company can take from it for overseeing its subsidiaries.
In companies, AI is often treated as a technical team project. 2026 surveys of directors suggest that is not enough: according to PwC's annual survey of directors, as reported by several summaries, a clear majority name AI as the skill their board most needs to strengthen, and only a minority put it on the agenda of every meeting. These figures concern US companies, often listed; they nonetheless describe a gap that applies to smaller structures too.
Why it rises to the level of governance
Three reasons make AI a governance subject rather than a mere tool. First, it engages accountability: a decision made or influenced by an automated system remains the company's decision. Second, it crosses internal boundaries: data, compliance, human resources, customer relations. Third, it evolves faster than existing frameworks: a use allowed in January can change nature with a new feature in April.
What a holding company can ask of its subsidiaries
For a company overseeing several entities, AI governance doesn't mean steering each tool. It means obtaining, regularly and comparably, a few simple answers:
- Where is AI used in each entity, and with what data?
- Who is accountable — a named person, not a diffuse team?
- Which uses are prohibited or subject to approval (customer data, decisions about people)?
- How are incidents escalated and handled?
- What does it cost, and how is the cost controlled?
A register rather than a rulebook
A twenty-page AI policy is rarely read. A short, current register — uses, owners, data involved, risk level — is, and it makes decisions traceable. It is also the best starting point for meeting regulatory requirements, notably those of the EU AI Act, whose calendar was adjusted by the Digital Omnibus but whose transparency obligations already apply.
Measure value, not just usage
Finally, industry analyses note that organizations getting real returns from AI are those where the subject comes back regularly before management, with outcome indicators rather than activity counters. A tool used massively but whose effect nobody measures is not a success: it is a cost.