When a Ledger Outlier Is Not an Error

Finance teams often ask for a detector that never misses. That request quietly assumes every outlier is wrong. In practice, legitimate seasonality, one-off restructures, and approved catch-up journals produce patterns that look irregular until you read the memo behind them.

Useful anomaly work starts by naming which accounts should be quiet, which should pulse with the calendar, and which exist only to absorb temporary imbalance. Without that map, a model—or a person—will escalate harmless activity and bury the few postings that matter.

We keep a short list of “expected irregularity” for each engagement: year-end bonuses, insurance renewals, deferred revenue true-ups. Flags that match those stories are demoted. Flags that break them rise.

The goal is not a clean dashboard. It is a short list of exceptions a controller can finish investigating before books close.