Plan, Do, Check, Act: The Loop That Runs Every Good System

Every management standard worth its salt runs on four words: Plan, Do, Check, Act. ISO/IEC 42001 is no exception.

Plan means deciding what your AI should do and what could go wrong before you switch it on. What is the system for? Who could it harm? What rules apply? You answer these on paper, not in a post-incident meeting.

Do means running the AI the way you planned, with trained people and clear controls. Not improvising. Following the plan you wrote when you were thinking clearly.

Check means measuring. Did the system behave the way you expected? Did anyone complain? Did the data drift? You look at the evidence on a schedule, not just when something breaks.

Act means fixing what the checking found. Update the plan, retrain the model or the people, tighten a control. Then the loop starts again.

Here is what I love about this loop: it assumes you will not get it right the first time. It was built for a world where conditions change. And nothing changes faster than AI.

Most AI failures I read about are not technology failures. They are loop failures. Someone planned poorly, or never checked, or checked and never acted.

Run the loop. It is that simple, and that hard.

I explain frameworks; I don't give legal counsel.