AI projects in logistics rarely fail because the technology “doesn’t work”. They fail because the organisation can’t explain, in operational terms, what an AI output is allowed to do. Sterdts has found that the fastest wins come from using AI to reduce coordination friction while keeping accountability with people. If an AI layer is trained on the clean slice only, it will be most confident precisely when the missing context matters. Control layers and approval gates: separating drafts from decisionsIn the sector generally, AI becomes risky when outputs slide from “draft support” into “implied promise” without a named owner.