Exceptions-only review means a person only handles the documents that failed a check. Clean, high-confidence waybills clear straight through and deliver their data automatically; the review queue holds just the exceptions — the ones with a failed validation or a low-confidence critical field. The work shrinks to the handful that genuinely need a human.
The alternative is checking everything: a person opens every waybill, re-reads every field, and rubber-stamps the ones that were fine to catch the few that weren't. That doesn't scale, and it's where attention runs out exactly when it matters. Exceptions-only inverts it — trust the checks to clear the clean documents, and spend human attention only where a rule failed.
What makes a document an exception
Two things route a document to review:
- A failed validation — an AWB check digit that doesn't resolve, a chargeable weight below both gross and volumetric, a consolidation that needs confirming.
- A low-confidence read on a critical field — the AWB number, the chargeable weight, a field the whole document turns on.
Everything else — having passed its checks with adequate confidence — clears.
With , you can let clean waybills clear and review only the exceptions.
Start hereWhy "clear everything" and "review everything" both fail
Auto-clearing everything ships wrong data downstream — a mis-billed charge or a rejected EDI message that surfaces days later. Reviewing everything drowns the real exceptions in routine confirmations and doesn't scale with volume. Exceptions-only is the middle path: a calibrated confidence threshold decides which documents clear and which get a human, so the queue length tracks the number of real problems rather than the number of documents.
# illustrative split — not a benchmark cleared = 44 # passed validation, confidence above threshold routed_to_review = 6 # 4 failed a check · 2 low-confidence critical field # a person sees 6, not 50
What it looks like in practice
Documents arrive by API or email, run through extract and validate, and split. The clean ones deliver their structured record automatically to wherever the data goes. The exceptions land in a review queue with the specific field and the reason it was flagged — so the reviewer fixes one field instead of re-reading the whole document. Approve or correct, and it rejoins the flow.