Canon · When the board asks
008

What boards actually ask

The capability demonstration is the wrong preparation. Boards ask four questions, none of them about what the technology can do, and a programme that has not rehearsed them will answer badly in public.

Updated May 2026
In brief
  • Boards are not evaluating whether the technology works. They assume it does. They are evaluating whether the organisation is exposed.
  • Four questions: what could leave the building, what you could be liable for, what it cost against what came back, and what happens if the vendor changes terms.
  • Bring the concentration number. Nobody asks for it directly and everybody in the room is calculating it.
  • A known gap is forgiven. A discovered one is not. Mark your weakest answer as weak before somebody finds it.

Programme teams prepare for board sessions by building a demonstration. It is the artefact they are proudest of and the one least likely to be asked for. Boards are not evaluating whether the technology works. They assume it works, roughly, because they read the same coverage everyone else does. They are evaluating whether the organisation is exposed.

The four questions

What could leave the building.Which data goes to which provider, under what contractual terms, retained for how long, and used for what. The answer must be specific to your contracts, not to the provider's public marketing. A board member who has read one article about training data will ask, and “they say they don't” is not an answer that survives follow-up.

What we could be liable for. Where output reaches a customer, a regulator or an employment decision without a human accountable by name. The useful preparation is a list of every deployment with the accountable human next to it, and an honest mark against the ones where that human is nominal.

What it costs and what came back. This is where the denominator discipline pays. A board that receives a return figure with its full cost base stated will trust the next figure. One that receives a licence-only number and later learns the real total will not trust anything from that programme again.

What happens if the vendor changes the terms. Price, model deprecation, regional availability, or acquisition. The question behind the question is whether you have built something you cannot get out of. Boards have watched this film before with other categories of supplier.

Exhibit 1
What the team prepares against what the room asks
The demonstration is the artefact the programme is proudest of and the one least likely to be requested.
Prepared
Asked
Opening
A capability demonstration
What could leave the building, and under what terms
Second
Model benchmarks
Where output reaches a customer or a regulator with no named human accountable
Third
Adoption curve
What it cost in full, and what came back
Fourth
Roadmap
What happens if the vendor changes price, deprecates a model, or is acquired
Unasked and calculated anyway
Nothing
Share of deployed value resting on a single provider

The preparation that works

  • One page per question, written before the meeting, with the weakest answer marked as weak. Boards forgive a known gap and punish a discovered one.
  • Bring the concentration number. What share of your deployed value depends on a single provider. Nobody asks for it directly and everyone is calculating it.
  • Have the exit answer ready even if it is unattractive. “Twelve weeks and a rebuild of the retrieval layer” is a real answer. “We would migrate” is not.

What to leave out

Model names, benchmark scores, and architecture diagrams. If a board member wants that detail they will ask, and the ones who ask usually want to establish that you know it rather than to hear it. Answer in one sentence and return to the exposure question they were really asking.