Purpose and fit

Use this to redesign a repeatable workflow before connecting tools or introducing automation. Start with the work as it happens today, including informal checks and exception handling.

Before you start

Use your existing process notes, SOP or an authorised description of one recent work item. Include the actual roles, systems, outputs and exceptions you know. If you do not have a formal process map, plain-language notes are enough for a first draft; unknown permissions or handoffs should stay unresolved.

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 task trigger, recipient and definition of completed work
  • Current steps, systems, source documents and role responsibilities
  • Common exceptions, missing-data cases and failure consequences
  • Existing permission, review, recordkeeping and escalation requirements

Practical workflow

  1. Walk through a recent, authorised example with the people who do the work. List the steps from request to accepted result.
  2. For each step, name the actor, required input and output. Mark whether the proposed AI role is read, suggest, draft or act.
  3. Define each handoff as a small contract: what is delivered, where it goes, how acceptance is checked and who accepts it.
  4. Add exception paths for missing or conflicting sources, a failed check, unavailable reviewers and tool failure. Each path needs a safe holding state or a known manual process.
  5. Rehearse a normal case, an ambiguous case and a failure case without live changes. Revise the map before a controlled trial.

Expected output

Expect a draft workflow map and handoff packet for a team walkthrough, not a working automation or verified access-control design.

A workflow map with one record per step:

Step/trigger | actor | authorised inputs | permitted action | output | recipient | acceptance check | approval needed | stop/escalation path | record kept

For every human handoff, include a short review packet:

  • The proposed output and the sources used
  • Unresolved questions or conflicting information
  • The checks already performed and their results
  • The decision required, its owner and the relevant deadline

Human checks

Frontline staff confirm that the map matches real work. The process owner checks that every transition has an owner and that work cannot silently bypass an unavailable reviewer. Use observable criteria such as a missing source or failed validation rather than relying on an AI system’s self-reported confidence. The system owner verifies actual access and action controls separately.

Limitations

A map is a design aid. It does not enforce approval, validate the software or discover every exception. A handoff that works in a rehearsal may need changes under real workload.

Starter prompt

Map the following workflow using only these notes: [insert approved workflow notes]. The intended outcome is [outcome]. Existing boundaries are [rules and roles].

For each step, identify the actor, input, permitted AI role, output, recipient, acceptance check and required approval. Mark missing information as “unresolved”. Add exception paths for missing sources, conflicting evidence, failed checks, unavailable reviewers and tool failure.

Produce a normal path and a concise handoff packet template. Flag any step with no accountable person or no safe stop. Suggest questions for a walkthrough with the team. Do not grant permissions, assume approvals or execute any step.

Return the normal path plus explicit exception paths, with one actor, input, output and acceptance owner per step where known. Mark unclear roles and approvals unresolved; show a safe hold or manual fallback rather than silently bypassing a missing reviewer. Do not execute the workflow.

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.