AI has moved onto the board agenda everywhere, but most board conversations still run on vendor talking points. Capital intensity in the AI economy is now extraordinary — Meta alone guided 2026 capital expenditure to $115–135 billion, and Microsoft's AI business passed $37 billion in annual run-rate — and the strategic question for every other company is how much of that wave to ride, and where. These twelve questions give directors a structure for the conversation that gets past the demo.
On Value
1. Which three business metrics will AI move this year, and by how much? If the answer is a list of pilots rather than metrics, the program is activity, not strategy.
2. What is our actual cost-to-serve change from AI so far? Ask for measured deltas with quality gates, not projected hours saved.
3. What would we stop funding if AI results don't materialize by Q4? Sunset criteria distinguish disciplined portfolios from science fairs.
On Risk
4. Where can an AI system take an action that harms a customer, and who reviews it? Agents act; scripts don't. The oversight model should name humans, thresholds, and rollback paths.
5. What is our vendor concentration, and what breaks if our main provider has a bad week? Every major model provider logged production incidents in the past year. Ask whether a tested failover exists, not whether one is planned.
6. Are we exposed to forced model migrations? Providers now retire models on hard deadlines measured in months. Someone should own the migration budget.
7. What data of ours can end up in someone else's model? Enterprise no-training defaults exist at every major vendor — verify they're in our contracts, including for tools employees adopted without procurement.
On Structure
8. Who owns AI results — and is it one person? Diffuse ownership produces diffuse results. Whether it's a Chief AI Officer or a line executive, the board should know the name.
9. What's our build-vs-buy line? Most companies should buy models and build the layer that touches their proprietary data and workflows. If management is training foundation models, ask why twice.
10. What is the talent plan beyond hiring? The binding constraint is usually existing staff who can deploy AI into existing processes, not ML researchers.
On Capital
11. How does our AI spend split between experiments, production, and platform? A healthy mix shifts toward production over time. All-experiments after two years is a red flag; so is zero experimentation.
12. If AI compresses our industry's cost structure by 30%, are we the beneficiary or the casualty? The only question on this list that's existential. It deserves a real session, not a slide.
How to Use These
Pick four per quarter and go deep rather than running all twelve as a checklist. The pattern in management's answers — measured vs. anecdotal, named owners vs. committees — tells the board more than any individual answer.
Internal Link Suggestions
The Chief AI Officer Role, Explained → aitj-chief-ai-officer-role
An AI Governance Framework for Enterprises → aitj-ai-governance-framework-enterprise
How to Measure the ROI of AI Agents → aitj-how-to-measure-ai-agent-roi
