DigitalCoa.ch

GBS AI Lab follow-up hub

← toolsDigitalCoa.ch

Tool 2

Preserving Judgment

Place each task where human judgment and AI execution belong.

Use this when the team needs a clear map of responsibility: what can be supported, what must stay manual, and what should never be automated without review.

Human Judgment+AI Execution
High Risk
Supported Execution
Manual
Low Risk

Purpose

  • Keep high-risk work under human review.
  • Prevent AI from being used where judgment should lead.
  • Create a clear responsibility map for the team.

When to use

  • The team is deciding which tasks belong to AI and which belong to people.
  • You need a shared map of responsibility for workflows.
  • There is a risk of over-automation or under-supervision.

When not to use

  • The task is already clearly manual.
  • The process is not yet mapped at all.
  • The team is looking for a shortcut rather than a decision aid.

Decision tree

Use the matrix to answer one question: where should judgment stay human?

  • If risk is high, keep judgment human.
  • If execution is supported but oversight matters, use the middle ground.
  • If risk is low and the task is structured, AI execution can help.

Process

01

Name the task

  • Write the workflow in plain language before assigning responsibility.

02

Test judgment and execution

  • Place the task on the matrix to clarify where humans must stay involved.

03

Define the handoff

  • Decide what AI may do, what humans must approve, and what stays manual.

AI prompts

Prompt 1

Map these tasks on a human judgment vs AI execution matrix. For each task, explain why it belongs in that quadrant.

Prompt 2

List the tasks that should remain human-led because the risk or judgment level is too high for AI execution.

Examples

Example

Student support emails may sit in supported execution, while disciplinary decisions stay in high-risk human judgment.

Common mistakes

  • Putting high-risk work into AI execution just because it is fast.
  • Treating every task as if it belongs in the same quadrant.
  • Ignoring the human approval step.

Downloadable checklist

  • Task named clearly.
  • Risk understood.
  • Judgment mapped.
  • Execution mapped.
  • Human approval defined.
Download checklist