A board approving next year's AI programme needs a clear account of the business commitments it is accepting. A list of products, pilots and training sessions will not establish whether the organisation can fund, control and operate them. Management should show what is already in use, which decisions AI can influence or execute and what evidence supports the proposed expansion.
The board's role is to set direction, oversee material risks and require an accountable management response. It should establish which decisions remain with the board and which it delegates within agreed limits. Routine configurations and individual case approvals belong with the people equipped to run the process.
These ten questions organise that discussion. Each calls for evidence and a decision rather than an assurance that AI is receiving attention. They are a proposed governance agenda for 2027, based on sources available through 30 September 2026; they are not a mandatory legal checklist or a forecast of next year's outcomes.
1. Which business outcomes justify our AI commitments?
Ask the chief executive to connect the leading proposals to a small number of business outcomes: service performance, completed work, capacity, risk reduction or commercial contribution. Require a baseline, a measurement owner and an explanation of why AI is a credible approach. Management should compare improving the existing process, configuring software already owned, buying and building.
The decision is which outcomes merit further commitment and which proposals lack a sufficient case. The board need not choose every implementation. It can require management to make comparable investment decisions within a portfolio envelope. The AI investment memo provides the workflow-level record behind that choice. An option to stop or defer should have a specific reason and a condition for reconsideration.
2. Where is AI already used, and how consequential is each use?
Request an inventory from management covering purchased tools, embedded capabilities, internal builds and external services. Give each material use an operating owner and distinguish advice, recommendations, record changes and external actions. A single total for “AI applications” conceals those differences. Procurement, technology and business teams should reconcile the inventory, including departmental purchases.
The MAS November 2025 proposed guidelines explicitly address identification of AI use, inventories and materiality assessments for financial institutions. In its 5 August 2026 reply, MAS described the guidelines as forthcoming. The board can require a dated inventory and proportionate reporting without assuming that proposal applies as a final rule to every business.
3. What authority are we willing to delegate to an agent?
A system drafting correspondence and one submitting a payment create different commitments. Ask the relevant executive to show which actions an agent may take, which systems enforce its limits and when a person must decide. Inspect one material workflow from input through approval to the resulting business record.
Singapore's January 2026 agentic-AI framework recommends bounding agents' powers and defining significant approval checkpoints. The board's decision is the risk appetite and escalation boundary for consequential authority; management designs and tests the controls within it. A supplier's description of a product as “autonomous” is insufficient evidence for granting it access. Any expansion of powers should trigger the appropriate management acceptance decision, with board escalation for material changes.
4. Who owns each material use, and can that owner control it?
Ask for the accountable business owner, deployment owner and incident lead for the most consequential uses. The business owner needs authority to narrow or suspend the system, a funded support arrangement and a clear escalation route. A committee can oversee the portfolio while a named executive remains responsible for a workflow's operating decision.
The March 2026 MindForge toolkit announcement identifies scope and oversight, risk management, lifecycle management and enabling capabilities as connected parts of implementation. Use those categories to challenge the management arrangement. The board should resolve ownership gaps and resource conflicts. Assigning a title to someone who lacks the capacity or authority to act does not settle the problem.
5. What evidence would change our decision to expand?
Require acceptance criteria agreed before the final evaluation, with results covering difficult cases and relevant user or customer groups. Management should distinguish correctly completed work from a model's isolated score and explain retained review, rework and serious errors. A finite evaluation with no observed serious failures leaves uncertainty; the permitted scope and monitoring should reflect it.
The decision is whether the evidence supports the requested commitment, warrants a narrower release or requires further evaluation. Technology and risk teams provide specialist challenge; the operating owner accepts the workflow within their authority. Ask for examples that failed and the resulting change in scope or design. The release evidence guide supplies a detailed record, allowing the board paper to focus on unresolved material issues.
6. What benefit reaches the business, and what does it cost to sustain?
Ask the chief financial officer to separate released staff capacity, avoided future hiring, realised spending reductions and revenue contribution. Count integration, supervision, exceptions, support and transition costs alongside licences and model usage. Benefits that depend on redeploying staff need a named destination for that capacity. Avoid counting the same hours as both a payroll reduction and a new service capability.
Require the downside case for the assumption most likely to change the investment decision, together with renewal and exit commitments. The board chooses the spending envelope and acceptable exposure; management owns the measurable benefit and review dates. The automation economics workbook supports the underlying comparison. Its illustrative calculations are decision aids, not evidence of returns already achieved.
7. How dependent are we on suppliers, and what survives their failure?
Ask the technology and operations executives to map the shared dependencies behind material workflows: models, hosting, identity, data and integration services. Separate products can depend on the same provider, creating an exposure that individual purchase approvals miss. Request what can be exported, how service changes are handled and which alternative arrangements have actually been tested.
The board's decision is which concentration and continuity risks it accepts, and what funded mitigation it requires. Replacing a supplier may need changed interfaces, fresh evaluation and staff retraining. A statement that data is portable does not establish that the business process can continue elsewhere. Management should quantify the interruption it can tolerate and identify where the proposed fallback remains unproven.
8. Which obligations and affected-party risks change across our markets?
A regional rollout needs a record for each entity covering data, customers, employees, decisions and relevant local requirements. Ask legal, risk and business owners to identify unresolved approvals and explain whether the operating arrangement works in each country. Singapore guidance offers a useful starting point; it does not decide another jurisdiction's requirements.
For Singapore personal data, the PDPC's 2024 AI advisory guidelines explain existing PDPA obligations and say organisations retain primary responsibility for ensuring their chosen system can meet those obligations. The decision is whether market-specific conditions justify release, a narrower scope or delay. Include how affected people can question or correct a material outcome, alongside the organisation's legal assessment.
9. Can we detect, contain and recover from a consequential failure?
Request a completed exercise involving the operating team, with evidence of the resulting business state. It should cover who can suspend authority, which actions may already have completed, how records are reconciled and how the team handles incoming work. Test fallback with the staff and suppliers that remain under the proposed budget.
The risk and operations executives should identify gaps that exceed agreed tolerance and recommend action. The board then accepts the residual exposure or requires remediation before further expansion. A working pause button alone does not establish recovery. Where a failure could harm a customer or counterparty, management should show the notification, escalation and remedy decisions it will need to make, without assuming every action can be reversed.
10. Which commitments will we stop, and what capacity remains for change?
Ask management to propose renewals, consolidation, closures and staged new investments in one portfolio. Include scarce specialist time as well as cash. Several fully funded projects can still compete for the same operations, security or data reviewers. Shared capabilities should serve identified workflow needs and have their own acceptance evidence.
The board should approve a sequenced portfolio with visible reserve, review dates and conditions for releasing further funding. The 2027 funding guide sets out the supporting method. Stopping an evaluation can be a sound decision when it resolves an important uncertainty; management should record that learning without manufacturing a financial return. Closure also needs money and ownership where contracts, records or business continuity remain.
Turn the agenda into a decision record
Ask the company secretary and responsible executive to organise the next AI paper around the decisions due, the supporting evidence, unresolved assumptions and accountable owners. Separate matters for board decision from management actions and information. For each material decision, record the permitted scope, conditions, review date and trigger for escalation.
Begin with the three uses that would matter most if they failed or absorbed more resources than expected. Require management to bring their inventory, investment case and operating evidence together. The board will then have a concrete basis for approving the programme, limiting its scope or asking management to return with the missing evidence.
Read our editorial principles for evidence, contribution and correction standards.