Ranking quality and review decisions answer different questions. ROC AUC compares positive-negative score orderings, while average precision summarizes precision as recall changes. Both use continuous event scores. You will compute each measure and inspect a precision-recall curve before placing either number in a business report. Average precision depends on the event fraction, so its reference level changes when the population changes. A hard classification removes most thresholds from the curve. Keep the positive label explicit, preserve array order, and remember that an attractive ranking score doesn't tell the operations team how many records it must review.