← Back to all guides

Process first. Then AI.

Capture how the work actually happens—its triggers, decisions, exceptions, and expected result—before choosing a prompt, AI assistant, or automation tool.

Why start with the process

Reliable AI implementation and automation do not start with selecting a tool or writing a prompt, but with understanding how the work actually happens.

  • Prevent automating errors: Starting with a specific tool often leads to automating an incorrect, inefficient, or redundant workflow.
  • Clear structure: A clear process map serves as a stable foundation for prompt design, automation rules, and review mechanisms.
  • Safe division of labor: Only a mapped workflow allows you to safely determine which steps belong to a human, where AI assists, and what a standard system handles.

What to map

Before designing any solution, capture these 6 key areas of your workflow:

  • What triggers the process? The exact input event (e.g., an incoming email, a chat message, a CRM status change, an uploaded file).
  • What is the final expected result? The required output (e.g., an approved estimate, a structured brief, a sent reply).
  • Who performs each step? Roles and responsibilities of specific team members or software systems.
  • Where do decisions, loops, and exceptions occur? Points requiring choices, missing details, or formal approvals.
  • What information is required for each decision? Input data, documents, business rules, and context needed for the correct step.
  • How do we verify the result is correct? Quality criteria, completeness checks, and security verification before handoff.

Current state and future state

When mapping a workflow, maintain a strict separation between two distinct documentation stages:

  • Current state (AS-IS): How the work happens today, including informal habits, manual handoffs, ambiguities, and undocumented edge cases.
  • Future state (TO-BE): How the process should operate after introducing AI assistance, automation rules, and streamlined steps.

Never merge both states into a single confusing diagram. Always capture today’s reality first before designing target changes.

Human, AI, and automation

Assign every step of the mapped workflow to one of three core solution layers:

1. Human

Expert judgment, approvals, final accountability for decisions, handling critical exceptions, and managing legal or ethical considerations.

2. AI

Text interpretation and summarization, extracting structured data from unstructured inputs, drafting initial responses, and classifying content.

3. Automation

Deterministic data transfers between systems, formatting, sending notifications, task routing, and scheduled status updates.

Simple working template

A structured template you can immediately use for a fast, clear workflow outline:

  1. Process Name
  2. Trigger
  3. Expected Result
  4. Participants and Responsibilities
  5. Current Process Steps
  6. Decisions and Exceptions
  7. Required Inputs
  8. Checkpoints
  9. What Remains with Human
  10. What AI Can Do
  11. What Automation Can Do
  12. Proposed Future Process
  13. How We Measure Value

When a simpler format is enough

You do not need complex documentation for every task. Choose the format based on complexity:

  • Simple task for one person: A concise step-by-step checklist is enough.
  • Detailed recurring procedure: A Standard Operating Procedure (SOP) structured as a clear guide.
  • Multiple people, systems, decision loops, and AI/automation: Build a clear process map or role flow diagram.

“First capture the work, decisions, and expected result. Only then decide where AI should help, where automation fits, and where a human must remain responsible.”

Related practical guides

Ready to continue with a practical AI use case?

Choose one of the practical guides below, or get in touch if you need help mapping your own process.

Choose a practical guide Need help mapping your process