I ran a fraud detection model on a municipal payroll dataset last week. Not the flashy AI kind—just pivot tables, variance thresholds, and a lot of coffee. The anomaly wasn’t in the big numbers, but in the third column of a recurring expense: 'office supplies' for a single employee who never used a printer. Turned out they were charging $120 for 'staples' every month for two years. I didn’t feel triumphant. Just tired. Like I’d caught someone doing something small and sad. Still, I sent the report. Someone else will handle the rest. That’s how it works.