Board & governance
Seven Board Questions About Your AI Strategy
By Kevin Williams · 4 min read
Published
A useful board conversation about AI connects business value, accountability, and evidence. Prepare to explain what the organization is trying to improve, what it has learned, and which decisions need oversight.
1. Which business outcome are we trying to improve?
Name the operating outcome before describing the technology. Revenue quality, service capacity, decision turnaround, and avoidable rework are possible areas to examine. “We are adopting AI” describes an activity, not a business result.
Show how the chosen workflow connects to that outcome. If the pilot only measures drafting speed, do not present it as proof of revenue growth. State where the evidence ends and what would need to be tested next.
2. Who owns the result and the consequences?
Name an accountable business owner and explain the supporting roles. The technical team may own system reliability, a functional leader may own workflow performance, and a qualified reviewer may control output acceptance. The board should be able to see where those responsibilities meet.
Avoid making “the AI team” responsible for every unresolved question. If the pilot changes customer commitments or employee work, the people responsible for those areas need an explicit role in the decision.
3. What evidence supports continuing?
Bring a comparison with the previous process and examples of both acceptable and unacceptable outputs. Describe the sample, the evaluation method, and the limitations. A single successful demonstration cannot establish how the workflow performs under ordinary operating conditions.
Separate reported time savings from realized business value. If saved minutes cannot be redeployed, they may still improve workload but should not automatically appear as cash savings in the business case.
4. What information and actions are allowed?
Explain what data enters the system, which applications are approved, and where human authorization remains necessary. Include how the team handles an exception or a suspected disclosure. The answer should describe an operating boundary people can follow.
NIST’s AI Risk Management Framework provides a voluntary structure for organizing AI risk work around Govern, Map, Measure, and Manage. It can help frame the discussion, but citing a framework does not establish that a particular deployment meets legal or contractual requirements.
5. What does the operating model really cost?
Include more than subscriptions or model usage. Integration, evaluation, review, maintenance, incident response, and support can all affect the economics. State which costs are measured, estimated, or still unknown.
Describe how costs may change with volume and with the number of exceptions. A workflow that is inexpensive during a supervised pilot can become costly when every department expects immediate support.
6. What changes for our people and customers?
Explain who will do different work, how they will learn it, and how the organization will collect feedback. Adoption depends on the workflow being usable, not just on employees receiving access to an application.
Give managers a way to report where the process fails without treating every objection as resistance. A user may be identifying a real exception that never appeared in the pilot. Customer-facing changes deserve the same attention to clarity, review, and escalation.
7. What would cause us to stop or change direction?
State the stop conditions before asking for expansion. They might include unacceptable error patterns, inability to control sensitive inputs, failure to improve the measured workflow, or review costs that outweigh the benefit. Tailor them to the actual use case.
A clear stop condition makes the strategy more credible. It shows that management is funding evidence and learning rather than defending a technology choice regardless of results. Include the fallback process so stopping does not leave the business without a way to operate.
How should you turn these answers into a board brief?
Use one page for the decision, the evidence, the risks, and the requested action. Put supporting material behind it. Clearly label a proposed pilot, a running experiment, and a production workflow so the board can tell what is actually operating.
Before the meeting, rehearse the question you would least like to receive. If the answer is uncertain, say what is unknown and who will resolve it. Coaching can help you make the uncertainty explicit without substituting confidence for evidence.
The next move
Bring the decision and the evidence to the board, not a catalogue of AI activity.