# Review one accepted invoice all the way to posting

Trace a record that bypassed manual review from its source page to the downstream payload. This reveals whether acceptance means more than a high extraction score.

By Cobnex editorial. Published 2026-09-10. Updated 2026-09-11.

## Choose a document with an adjustment

Use a synthetic invoice containing a credit, deposit or other explicit adjustment. Ask the team to identify the printed payable amount and the fields used by the validator.

Compare raw text, normalised values and page locations. Check whether currency, signs and decimal handling remain correct. A field can be recognised accurately and still be transformed incorrectly before posting.

Confirm the supported calculation for that document class. Do not accept a generic formula that ignores the adjustment simply because most demonstration invoices are simpler.

## Inspect the acceptance decision

Find the rule version and evidence that allowed the record to proceed automatically. Separate confidence thresholds, required-field checks, arithmetic and duplicate detection.

Ask how each rule was evaluated on reviewed documents, including accepted records. Reviewing only exceptions cannot establish the quality of automatic acceptance.

Look for conditions represented as unknown. Missing currency, unavailable confidence or an unrecognised document class should not become optimistic defaults without an explicit business rule.

## Follow the exact downstream payload

Compare the accepted record with the submitted values and attachments. Verify that a stale extraction result or unreviewed field cannot replace a corrected value during mapping.

Inspect operation identity and retry behaviour. A timeout after posting should lead to reconciliation or an idempotent retry, not a new independent invoice submission.

Check supplier and target organisation selection. Correct amounts posted to the wrong account remain a serious application failure that extraction metrics may not reveal.

## Require an end-to-end exception case

Run a fixture with a plausible but wrong payable field and confirm it is rejected or reviewed according to the acceptance rule. Inspect the source highlight and correction history, then verify the corrected downstream result.

Record known document classes outside automatic acceptance and the process for adding them. The review should establish a traceable chain from printed evidence to accepted business record, with explicit handling for contradictions and uncertain submissions.

## Sources

- [Microsoft Learn: extraction confidence](https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/concept/accuracy-confidence?view=doc-intel-4.0.0)
- [AWS Builders' Library: idempotent APIs](https://aws.amazon.com/builders-library/making-retries-safe-with-idempotent-APIs/)