Human validation tier

The human validation tier is an optional step where a reviewer checks low-confidence and edge-case fields after AI extraction — before the document enters your approval flow.

What it catches

Confident-but-wrong fields. The AI scores a field at high confidence, but the value is genuinely incorrect — a transposed digit in a bank account, the wrong currency code, or an amount read from the wrong column. These would pass silently without human review.
Edge cases in line capture. One wrong line can break order matching for the entire document. Human review makes line data reliable enough for automatic PO matching.
Non-standard layouts. Scanned PDFs, foreign-language inserts, and unusual invoice formats score lower confidence and benefit most from review.

Why it matters

Without validation, a high-confidence wrong field reaches approvers with no warning. With validation enabled, reviewers correct those fields first, so approvers can trust the pre-filled data and focus on approval decisions rather than data checks.

When it may not be needed

Validation adds a small processing delay and a per-document cost. It may not be cost-effective for clients with very low document volumes, highly standardised invoices from a small fixed supplier set, or clients who prefer their own bookkeeper to review every captured field.