LIWLuttrell Intelligence Works

checklist

Workflow Classification Worksheet

Estimated reading time
3 min read
Reviewed
Reviewed Aug 18, 2026

A worksheet for sorting workflow steps into AI-can-do, AI-can-draft, and human-only categories with review gates.

Type
Checklist
Audience
Operator
Level
Beginner
Format
Worksheet

Use this resource

Prints without site navigation for working sessions, review packets, or handoff notes.

Workflow Classification Worksheet · Reviewed Aug 18, 2026 · Luttrell Intelligence Works

Use this worksheet when a team is deciding where AI belongs in a real workflow. It helps separate routine support from draft work that needs review and decisions that must stay human-owned.

The goal is not to use AI everywhere. The goal is to make one workflow clear enough that people can see which steps are safe to assist, which steps need review, and which steps should not be delegated.

When To Use It

Use this worksheet when:

  • a team is already experimenting with AI informally
  • a workflow has repeated drafting, summarizing, organizing, or review work
  • people disagree about whether AI is safe for a task
  • the team needs a pilot candidate with explicit guardrails

Do not use it as approval to automate a workflow. Use it to decide what a safe first pilot might be.

Sample Workflow

If the team does not have a workflow ready, practice with a generic weekly project status workflow:

  1. Gather updates from tickets, meeting notes, and recent deliverables.
  2. Identify completed work, active risks, blockers, and next decisions.
  3. Draft a status summary.
  4. Verify links, owners, dates, and open risks.
  5. Send the final status update to stakeholders.

Step 1: Map The Workflow

List the repeated steps in the workflow. For each step, capture:

  • source inputs
  • who performs it today
  • how often it happens
  • what can go wrong
  • who trusts or consumes the output

Workflow name:

Current owner:

Primary receiver:

Steps:

  • Step:
    • Inputs:
    • Current owner:
    • Frequency:
    • What can go wrong:
    • Who relies on it:

Step 2: Classify Each Step

Choose one category for each step.

AI Can Do

Use this category for low-risk transformation, formatting, extraction, sorting, dedupe, or first-pass aggregation where deterministic checks or close human spot checks are enough.

Examples:

  • format rough notes into a consistent outline
  • dedupe a list of repeated items
  • sort source excerpts by topic
  • extract dates that a person will verify

AI Can Draft

Use this category for summarization, synthesis, first-pass communication, candidate decisions, recommendations, or comparison where judgment still matters.

Examples:

  • draft a status summary from reviewed source notes
  • propose test cases from acceptance criteria
  • summarize risks for a project owner
  • compare two options against a checklist

Human Only

Use this category for accountable decisions, external commitments, sensitive approvals, personnel matters, production changes, or anything where the organization cannot delegate responsibility.

Examples:

  • approve a stakeholder-facing message
  • decide whether a risk is acceptable
  • make a production change
  • commit budget, scope, staffing, or policy

Classification Grid

Copy one row per workflow step.

Step:
Category: [AI can do / AI can draft / human only]
Why this category:
Allowed source material:
Required review:
Evidence to keep:
Stop or escalation trigger:

Step 3: Add Review Gates

For every AI can draft step, define:

  • who reviews the draft
  • what they check
  • what evidence they need
  • what the AI is not allowed to do
  • when the work must stop and escalate

Review gate:

Draft step:
Reviewer:
Reviewer checks:
Evidence required:
AI must not:
Escalate when:

Step 4: Pick The Pilot

Choose one candidate pilot. A good first pilot is narrow, reversible, reviewable, and useful even if it stays in draft mode.

Pilot candidate:

Why this is a good first test:

People involved:

Approved sources:

Review cadence:

Success signal:

Stop condition:

Output Card

Summarize the workshop output in one page.

Workflow:
Pilot candidate:
AI-can-do steps:
AI-can-draft steps:
Human-only steps:
Allowed sources:
Review and approval rules:
Evidence checklist:
Owner:
Next decision:

Minimum Guardrails

Before piloting, confirm:

  • no private or customer-specific examples are used in public training material
  • no external sends or production actions happen automatically
  • AI output cites or links the sources it used where practical
  • stakeholder-facing output still requires explicit human approval
  • the team has defined what a failed output looks like before the pilot starts

Final Decision

Choose one:

  • Ready for a small reviewed pilot.
  • Needs narrower scope.
  • Needs better source material.
  • Needs clearer review ownership.
  • Not appropriate for AI assistance right now.

Decision:

Reason:

Next owner: