A high field score can hide unusable documents. This hypothetical experiment asks whether a complete invoice record is ready for its next reviewer. It does not authorize payment.

Start with one hundred held-out invoices: sixty familiar layouts, twenty-five new supplier layouts, and fifteen imperfect scans. Agree the required fields and matching rules before inspecting results.

Suppose four hundred eighty of five hundred fields are correct. That is ninety-six percent. Yet only eighty-four of one hundred records are completely correct. Errors cluster, so count whole records directly.

Now test a frozen routing rule. Seventy records move forward, including two incorrect ones. Thirty are held, including fourteen incorrect ones. The matrix reveals both escaped errors and review demand.

Check how serious the errors are and whether the team can review thirty records on time. Keep failed documents in the denominator. These are illustrative counts, not a benchmark. Expand the test before expanding the workflow.