Random samples impress textbooks. Operational desks usually need something else: a risk-weighted pull that still admits what it did not cover.
Stratify first
Split populations by amount bands, new vendors, or manual journals. Sample more heavily where override risk is highest. Record the strata in the working paper so a reviewer does not assume you cherry-picked favorites.
Write stop-or-go rules
Decide in advance when findings expand the sample. Without that rule, every exception becomes a political negotiation mid-close. Our flagship course includes a worksheet for these thresholds.
Confess coverage
State what percent of value and of count you tested. Incomplete coverage is acceptable when the limitation is explicit. Hidden coverage is what damages trust.