Every exception has an owner
An alert with no owner is noise. Before a test goes live, we agree who receives its exceptions and what they must do. A duplicate invoice goes to the AP lead, who confirms it was not paid twice or recovers the payment. A journal entry posted and approved by the same person goes to the controller, who reviews it and signs it off. A leaver still active in the bank portal goes to whoever administers the portal, who removes the user that day. Each exception closes with a reason code and evidence, and anything open past its due date appears by name in the monthly report.
A person reviews before routing
Raw test output always contains false positives: a weekend entry that was a scheduled accrual, a round payment that was a contractual retainer. A Finance Ethics analyst reviews every exception before it is routed. Expected items are documented and suppressed by rule. Real ones go to the owner with a one-line explanation. That review is the difference between monitoring people read and monitoring people mute.
Thresholds are tuned, not guessed
During setup we run every test over the past twelve months to see what it would have flagged. If a just-below-limit test would have fired forty times a month, the threshold or the test is wrong. The quarterly review keeps tuning, because new vendors, new systems, a new approval limit or a new AI agent in AP each change what should be tested.
The archive is built for auditors
Every test run is stored with its date, data source, parameters and result, and every exception with its owner, reason code, evidence and closing date. When an auditor, lender or acquirer asks how a control operated in a given quarter, the answer is a filtered export, not a reconstruction from inboxes. If an owner disputes an exception, the dispute and its outcome are kept in the record as well.
Results can back public statements
If you publish an Integrity Center, monitoring results support its attestations. When the dual-release test fails, the officer who signed the dual-release statement knows within the same cycle. Monitoring often begins after a one-off full-ledger analysis, which clears out the historic problems so the weekly tests start from a clean baseline. For the tests most programs start with, see continuous controls monitoring: ten automated tests.
