Ask a vendor how good their AI is and you get a percentage. That number is usually true, and it rarely tells you whether the tool belongs in case processing. It doesn’t show which errors occur, it misses fields the tool silently drops, and it says nothing about the expert time still needed before an output is safe to use. This talk introduces PHAIR, the Pharmaceutical AI Risk & Readiness Framework. PHAIR assesses the use, not the product: a ten-minute triage, a two-axis risk matrix with a detectability modifier for omissions, non-compensable control gates, and Human Correction Burden as a condition of approval. A worked PV example shows how the same tool can earn a GO for one job and a NO-GO for another.