Purpose and fit

Use this when an AI output informs a decision that someone must explain, approve or revisit. Make the AI contribution explicit and preserve the decision rights already established by the organisation.

Before you start

Use your existing decision brief, approval policy and role descriptions. Provide only the approved evidence summary needed to structure the record. An incomplete authority map is acceptable input: indicate which roles are known and which need confirmation instead of inventing names or approvals.

Use only information you are authorised to share with the chosen AI tool. Adjust the prompt to the fields and context you actually have; a specially created dataset is not required.

Required inputs

  • The decision, its scope, affected people and possible consequences
  • Existing authority, review and escalation requirements
  • Evidence the decision needs and the sources available
  • The proposed AI role, such as organising information or drafting options

Practical workflow

  1. State the decision in plain language and identify the person with authority to make it. Resolve an ownership gap before relying on the output.
  2. Separate preparation, evidence checking, advice, approval and execution. Assign the required people and confirm they accept those responsibilities.
  3. Specify the AI contribution and the evidence reviewers must inspect directly. Identify decisions or data that require specialist review.
  4. Prepare a proportionate decision packet: options, evidence, uncertainties, consequences and required approvals.
  5. The decision owner records the decision, rationale and conditions. Name the triggers that would require it to be revisited and the path for concerns or correction.

Expected output

Expect a draft accountability map and decision-record structure, not a recommendation on the merits of a consequential decision or a grant of authority.

An accountability map and a decision record:

  • Decision and scope
  • Decision owner and basis of authority
  • Preparer, domain reviewer, required approver and execution owner
  • AI task and relevant tool/version, where material
  • Evidence inspected, source references and unresolved uncertainty
  • Options considered and material consequences
  • Decision, rationale, conditions and actual approval record
  • Review trigger, correction/escalation route and record location

Human checks

The named people confirm their roles; listing a name is not consent or approval. Reviewers verify important evidence rather than accepting a generated citation. The decision owner can question, change or reject the recommendation and has enough time and information to do so. Follow established specialist and authorisation requirements for consequential decisions.

Limitations

A responsibility map does not create authority or transfer accountability to software. A completed record does not prove that a decision is sound or compliant. The level of review must reflect the task’s consequences and the organisation’s rules.

Starter prompt

Help structure an accountability map for [decision]. Existing decision rights and required reviews are [rules]. The proposed AI contribution is [contribution]. Available evidence is [approved summary].

Separate preparation, evidence checking, advice, decision approval and execution. Identify any missing owner, conflicting responsibility or unresolved approval requirement. Draft a decision record with source references, uncertainty, consequences and reconsideration triggers. Use “unassigned” or “not recorded” where information is missing. Do not assign authority, assert that anyone approved, or make the final decision.

Separate the AI preparation role from evidence review, approval and execution. Reference supplied authority rules and evidence; mark gaps unassigned or not recorded. List conflicts and reviewer questions before any decision fields. Leave the actual decision and approvals blank unless supplied as recorded facts.

Before drafting, ask a focused question if a missing detail materially changes the task or acceptance criteria. Otherwise produce a clearly limited draft and list the gaps. Label any non-critical assumptions; never invent facts, figures, sources, approvals or completed actions.

Data and AI output disclaimer

Use relevant data that exists or that you provide, with accuracy and completeness sufficient for the task. Different AI models and versions may produce different results, and responses may vary between runs. You are responsible for checking and validating the output against your own requirements, source information and acceptance criteria before relying on it.