playbook · 8 min read

How to choose the first AI POC

Compress a large ambition into one testable unit using value, feasibility, risk and adoption complexity.

Updated 2026-08-10

A POC is not a smaller version of a big system or a quick model demo. It should answer one clear question: if AI enters one action, will the improvement be observable, repeatable and judgeable by the owner?

First remove high-risk, low-visibility and ownerless processes. Prefer read-only, human-reviewed and reversible steps.

Write the validation plan before starting: baseline, evaluation data, owner, stop rule and expansion condition. Without it, a POC becomes an unexplainable demo.