A management team leaves a board meeting believing it has permission to expand an AI service. Several directors leave believing they supported further development, subject to evidence still to come. The minute says the proposal was approved.
That is a hypothetical situation. It illustrates a problem worth resolving before the meeting closes: a record of agreement can leave the actual decision uncertain.
For a board chair, the useful question is specific. What changes because of this decision, and what remains management's responsibility? Answering it gives directors a clearer proposal to judge and management a clearer basis for its next action.
Give the proposal a defined decision
Consider a hypothetical supplier service that reads correspondence and prepares order amendments. Management wants to extend the service into another business unit and allow it to submit certain amendments. The current workflow has been tested. The proposed unit uses a different submission path, and its fallback has not yet been exercised. For this example, assume the company's existing governance arrangements reserve approval of this expansion to the board. Management remains responsible for operating the current service within its existing authority.
Several matters could legitimately reach the board. Directors might be considering a material investment, reviewing management's approach to a significant exposure, or deciding a matter specifically reserved under the company's governance arrangements. Each requires a different conclusion. None is adequately described by the agenda label “AI approval.”
Before presenting the recommendation, the sponsor should explain what the board is being asked to decide, why the matter belongs there and what the decision would permit. If the request is for oversight rather than authorization, say so. Where the mandate is unclear, use the company's established governance and legal route to resolve it.
The G20/OECD Principles of Corporate Governance describe boards' strategic guidance and monitoring of management, including risk management and reporting integrity. They also recognize differences in board structures and jurisdictions. They do not determine the legal duties or reserved decisions of an individual company.[1]
Touch Stone's recommendation is to preserve that distinction in the request itself. The board evaluates the matter within its actual mandate. Management owns its operating recommendation, the implementation and the evidence supporting it. A director can ask detailed questions about a consequential control without becoming responsible for designing that control.
Keep the alternative visible
The supplier-service proposal should explain what happens if the requested expansion does not proceed. Management might retain the tested workflow, continue preparation assistance while people submit amendments, or return with a narrower proposal.
Those alternatives have consequences. Manual submission may consume staff time. Deferral may delay a benefit management expects. A narrower scope may leave part of the service unchanged. Management should describe those consequences without treating an expected benefit as an observed result.
An explicit alternative makes the decision more useful. Directors can compare the recommendation with a feasible operating choice. They can also distinguish a temporary decision to retain the current service from an indefinite rejection of the project.
In this example, evidence about the existing submission path does not, by itself, establish that the proposed path will work. The board can recognize the value of the current service while asking management to address the expansion gap. That judgment need not imply a failure in the current operation.
Make a condition usable
“Approved subject to satisfactory testing” leaves several questions open. Which testing? What result is satisfactory? Who assesses it? Does the decision permit expansion immediately, or only after a defined condition is met?
A useful condition identifies the scope it affects and the event that changes the decision's effect. Management needs a way to observe and enforce it. If a condition depends on a vendor record that the company cannot obtain, the condition may be impossible to satisfy through the proposed route.
For the hypothetical supplier service, the board could defer approval of expansion and ask management to return with evidence about the new fallback. Alternatively, management might present a separable extension that does not rely on that path and explain why the available evidence applies. The assumed reserved decision applies only to this constructed case; another company must establish its own mandate before choosing its response.
The record should retain the reason for the chosen posture. “Further evidence required” is less useful than identifying the unresolved dependency and why it matters to the proposed scope. Management then knows what would make the proposal ready for another decision.
Read the minute as an operating handoff
The company secretary and relevant advisers can fit this discipline to the organization's existing records. There is no need to invent a parallel approval system or treat a suggested working aid as a mandatory template.
Before the record is finalized, read it against the recommendation. Does it preserve the specific matter decided, the scope affected and the material qualification? Does it identify the next management or assurance contribution where one remains necessary? Could someone reasonably read it as certifying an implementation that the board did not examine?
A compact, hypothetical account of the supplier discussion could record that the board deferred approval of the proposed expansion because evidence for the additional submission path remained incomplete. Expansion would not proceed under that decision. Management would return with the specified fallback evidence for reconsideration and remain responsible for the current service within its existing authority; the decision would not extend that authority or certify the current implementation. This is an illustration of the information to preserve, not model legal wording or a resolution for adoption.
The most useful check is to give the record to a colleague who did not attend. Ask that person to explain what can happen next and what still needs to be established. If the answers differ from the chair's and sponsor's understanding, resolve the ambiguity before it becomes the reference for implementation.
Preserve the reason when circumstances change
A decision can be reasonable when made and require reconsideration later. The provider may change a path on which the recommendation depended. A receiving team may lose the capacity assumed in the proposal. Evidence that was available may become unavailable.
The record should make the decisive assumptions visible enough for management and the appropriate governance authority to recognize a relevant change. This does not mean every update requires another board vote. The significance of the change and the company's existing allocation of responsibility determine the route.
For the next AI proposal, begin with one sentence from the management sponsor: “We are asking this body to decide the following matter.” Then ask the chair and secretary whether the proposed decision record says the same thing.
The goal is a decision whose meaning survives the meeting. Directors should be able to explain their judgment. Management should know what it can do next, which questions remain open and how to return with an answer.
Source and scope
[1] OECD, G20/OECD Principles of Corporate Governance 2023: Responsibilities of the board, 2023, Chapter V. These principles recognize structural and jurisdictional differences; they are not a determination of a particular director's legal duties.
The supplier-service situation is hypothetical. The proposed decision-record practices are Touch Stone Publishers' editorial recommendations, not research findings that demonstrate an outcome or legal requirements.
For a discussion of the governance questions facing your organization, explore Touch Stone Publishers engagements.