The trigger
A board or executive committee paper is due and the item is a genuine decision with more than one live option. Boards are expected to ask where the numbers came from and what residual risk remains; a paper that cannot answer fails before it is tabled.
The inputs
- The decision question in one sentence, with the named decision-maker and deadline
- The options considered so far, including doing nothing
- The evidence: reports, register extracts and figures, each with its source named
- Constraints (budget, headcount, policy, timing) and stakeholders affected
- The decision-memo skill file installed in the approved AI workspace
- [ORGANISATION]'s board paper template and AI-use disclosure norm
The steps
- Restate the decision as one sharp question with the deadline attached; split anything that is really two decisions.
- Assemble the evidence pack to register discipline: every figure traceable to a named source, current, with an accountable owner. Anything failing that test stays out and the gap is named.
- Run the first prompt against the skill file, supplying only the framed question, options, constraints and sourced evidence.
- Rework the returned memo into [ORGANISATION]'s board paper template, keeping each figure's source beside it and the decision-sought line intact.
- Run the caveat-preservation pass with the second prompt, comparing the paper against the source material.
- Restore each missing qualifier, or record its omission as a deliberate authorial choice.
- Finalise the decision-sought line, naming [DECISION_MAKER] and the date, and submit the paper to the pack.
The prompts
Run the skill file first, then the QA pass on the reworked paper.
The checks
- Every figure appears in the supplied evidence with its source named; nothing was invented to fill a gap.
- The do-nothing option is present with consequences, and evidence gaps are stated plainly.
- Every caveat-list item is restored or recorded as deliberate, because a summary that keeps the number and drops the qualifier changes the decision.
- The recommendation still reads as a draft to test; the decision-sought line names who decides and by when.
- Dates and figures are re-confirmed against the source records.
Governance controls
- The author owns every sentence and answers the board's questions without the tool; the paper supports the decision, [DECISION_MAKER] makes it.
- Nothing board-confidential enters a tool [ORGANISATION] has not approved; if material appears sensitive, stop and confirm the workspace first.
- AI-assisted drafting is disclosed per [ORGANISATION]'s norms for board material.
- Any caveat deliberately dropped is a recorded authorial choice, never a silent compression.
- Accountability never moves to the model: the tool never states the final decision.
The output
A decision-ready board paper in [ORGANISATION]'s template, with options including doing nothing, every source named, surviving caveats, a draft recommendation and a decision-sought line, filed to the board pack with its source material.
Time it takes
About two hours the first run, mostly installing the skill file and assembling evidence. A practised run fits ninety minutes, with evidence assembly and the caveat pass taking the largest shares.
TheAICommand. Intelligence, At Your Command.