When a Document Has to Pass Through Five Systems

The stages

1. Intake — email, portal, scan or upload, all normalised to one queue
2. Classification — what kind of document is this
3. Extraction — the fields, per document type
4. Validation — against master data and internal consistency
5. Review — for anything uncertain or consequential
6. Write-back — into the destination system, idempotently

One place holds the state

Every item has a record with its current stage, its history and its data. Without that, a document that stalls between stages is invisible, and invisible stalls are how trust dies.

That record is also what makes the audit trail possible and what lets a person answer “where is my invoice?” in five seconds.

Each stage independently retryable

A failure at validation should not require re-extracting. Store the output of each stage so a retry starts from the last good point.

It saves cost, it saves time, and it makes incidents far less dramatic.

Design the queues properly

• Ordered by consequence rather than arrival
• Age visible and alerted on, so nothing sits unnoticed
• A quarantine for items that failed repeatedly, with the reason
• A way to reprocess a quarantined item after a fix

Reconcile daily

Compare what entered intake against what reached the destination, every day. The difference should be explainable in one line: in review, in quarantine, or rejected.

Any unexplained gap is a silent failure, and finding it the same day is the whole point.