Trust & Human ControlGuide

When Not to Automate a Business Decision

Identify decisions that should remain under meaningful human control because of accountability, uncertainty, rights or irreversible consequences.

Published:
Direct answer

Do not automate final authority when consequences are material, evidence is incomplete, affected people need recourse, policy requires accountable judgement or the decision cannot be safely reversed.

Practical method

Five stop signals

A stop signal does not forbid useful automation around the decision. It defines where a person must understand, challenge and approve.

  1. 01

    Material harm

    The choice can materially affect rights, safety, employment, access or a major financial commitment.

  2. 02

    Unresolved uncertainty

    Key inputs are assumptions or unknowns and the system cannot surface that uncertainty clearly.

  3. 03

    No effective recourse

    Affected people cannot correct data, challenge reasoning or reach an accountable reviewer.

  4. 04

    Irreversibility

    The action is difficult to undo before harm occurs.

  5. 05

    Contextual judgement

    Policy intentionally requires values, proportionality or exceptions that cannot be reduced to stable rules.

Worked example

Automate preparation, not authority

A system may collect supplier evidence, flag missing declarations and compare responses. A named committee can still review conflicts, consider exceptions and approve the selection. The useful boundary is between repeatable preparation and accountable judgement.

Limits and safeguards

What this method cannot guarantee

  • Human review must be meaningful; a reflexive click is not a control.
  • Keeping a person in the loop does not correct biased evidence or poor policy by itself.
  • The method supports judgement; it does not transfer accountability from the people approving the decision.