When a network mixes high-volume city branches with quieter satellite sites, a flat random sample often over-weights the biggest location and under-examines the rest. We start by grouping branches into strata by monthly expense spend and claim count.
Inside each stratum we pull lines from travel, entertainment, and miscellaneous reimbursements in proportions that reflect local risk—not only local volume. A small branch that routinely overrides approval thresholds can deserve denser sampling than a large site with clean dual control.
Cut-off timing matters. Claims posted in the final week of a quarter often hide incomplete vouchers pushed through to meet reporting deadlines. We earmark those days for extra selection density.
The output is not a statistical certificate; it is a defensible sample that your finance committee can explain to external auditors and to branch managers who ask why their site was reviewed.