The questions every board should ask before approving an AI investment
Boards should treat AI proposals as strategic change programmes, not software purchases. The right questions probe use case clarity, governance, risk, capability and operating discipline. Research suggests AI value depends on deliberate management, while capability and responsible-use practices are advancing at different speeds.
AI investment papers often arrive framed as inevitabilities: competitors are moving, productivity gains are available, and the organisation must not fall behind. That pressure can make boards approve too quickly. The better test is simpler and harder: what problem is this solving, what new risk is it introducing, and can the organisation govern it well enough to realise value? Evidence from university research points to a pattern. AI capability is progressing quickly, but responsible-use practices are less consistently measured; organisations also tend to see value only when implementation is matched to clear governance and user support 13. Boards therefore need questions that separate excitement from readiness, and ambition from control.
Start with the business problem, not the technology
The first board question should be whether the proposed investment is anchored to a specific business issue. MIT Sloan’s research distinguishes broadly applicable tools, which lift individual productivity, from tailored solutions built for a defined group and use case 3. Those are different investments with different risks, operating models and payback assumptions. A board should ask: what decision, process or customer interaction will change if this is approved? If leaders cannot explain the working problem in plain terms, the proposal is probably too vague to fund.
Boards should also ask what would happen if the organisation did nothing for a little longer. That question is not about delaying innovation; it is about forcing prioritisation. If the benefit is convenience rather than strategic advantage, the case may belong in a smaller pilot rather than enterprise approval. By contrast, if the problem is tied to service quality, compliance workload, knowledge retrieval or coordination, the board can examine the case against a clearer operating need.
Test whether the organisation can govern the use case
The governance question is not simply whether there is a policy. It is whether the organisation can keep pace with the way the tool will actually be used. MIT Sloan notes risks around unvetted tools, data loss, intellectual property leakage, copyright issues and security breaches, and recommends cross-functional guidelines involving technology, legal, privacy and governance 3. Boards should ask who owns those controls, how exceptions are approved and how shadow use will be detected. If the answer relies on informal vigilance, the investment is not yet board-ready.
A second governance question concerns model choice and vendor dependence. For tailored solutions, MIT Sloan highlights trade-offs between buying, building or enhancing models, with implications for transparency, context and long-term cost 3. Boards should ask what sits behind the interface, where data goes, whether the vendor can explain failure modes, and how exit would work if performance, pricing or terms change. These are not technical curiosities; they shape resilience and bargaining power.
Probe the risk, assurance and measurement plan
Stanford’s AI Index shows strong progress in capability, but also notes that responsible-AI reporting remains patchier than capability reporting and that documented incidents have risen 1. Boards should not treat that as a reason to avoid AI; they should treat it as a reason to insist on assurance. Before approving spend, ask what will be tested, by whom, and at what points in the lifecycle. The board should want evidence of red-teaming, human review thresholds, escalation paths and clear rules for when the system must not be used.
The measurement plan matters as much as the model. Boards should ask how success will be judged beyond adoption. Useful metrics are likely to include process quality, error rates, cycle time, user confidence, customer impact and control effectiveness. If the business case only counts potential savings but not the cost of supervision, retraining and exception handling, the case is incomplete. The practical test is whether management can show that benefits are realisable under ordinary operating conditions, not only in a demonstration environment.
Check leadership, capability and adoption readiness
Harvard Business School research on leadership with AI agents suggests that effective leaders ask more questions, create turn-taking and reassess as they go 2. For boards, that is a useful reminder that AI adoption is not mainly a procurement exercise; it is a coordination exercise. The relevant question is whether leaders can create enough curiosity and discipline for people to use the tool well. If the proposal assumes behaviour change without training, time or managerial support, uptake may be superficial.
Boards should ask who will translate the investment into daily practice. That includes defining acceptable use, designing training for specific roles, and deciding how managers will reinforce good habits. It is also worth asking whether the organisation has enough internal capability to challenge vendors and interpret outputs. A strong business case includes line leaders, risk functions and operational users, not just IT. In our view at Authenticate, boards are best served when AI proposals are assessed as change programmes with named owners, not as standalone systems.
Ask whether the organisation is ready for the human consequences
MIT Sloan’s work on AI leadership stresses that trust, transparency and a coherent narrative matter when organisations introduce generative AI 4. Boards should ask how employees will understand the change, what worries have been anticipated, and how the organisation will avoid normalising mistrust. This is especially important where AI touches performance assessment, scheduling, employee communications or knowledge work. If people do not understand how outputs are produced, they may either over-trust the system or ignore it entirely.
There is also a broader workforce question: which tasks are being augmented, which are being automated, and what is the transition plan for people whose work changes? Boards do not need every answer on day one, but they do need evidence that management has thought beyond implementation. A cautious, staged approach is usually preferable to a wide launch with weak support. The board’s role is to insist that enthusiasm is matched by consent, clarity and accountability.
Sources
- The 2026 AI Index ReportStanford University — Human-Centered AI · University research · link checked 15 October 2026
- Think You’re a Good Leader? Try Leading an AI TeamHarvard Business School AI Institute · University research · link checked 15 October 2026
- How to manage two types of generative artificial intelligenceMIT Sloan School of Management · University research · link checked 15 October 2026
- Who should be the head of generative AI — and what they should doMIT Sloan School of Management · University research · link checked 15 October 2026
Further professional reading
External professional resources, separate from the numbered sources above. These are not university studies, and listing them does not indicate any collaboration, certification, affiliation or endorsement.
- Global Institute of Directors (opens in a new window)External professional resource · governance and directorship
- World Coaching Institute (opens in a new window)External professional resource · coaching practice