Human Review used to fire the moment OCR finished. There was no gap to fix anything first, so reviewers opened whatever OCR pulled out: formatting problems, missing fields, and all. They spent half their time correcting text instead of checking whether the underlying data was actually right. Enable Human Review changes that. It's a new action that sits between OCR and the review step, so you can clean up the data before a reviewer ever sees it.

How it works

A document still runs through Split OCR the same way it always has. What's changed is that Human Review no longer fires automatically right after. It's off by default.

That gap is where your own logic goes. Drop in code or business rules to normalize a number, fill in a missing field, or apply a default, whatever the data needs. Once that's done, turn on Enable Human Review, and the reviewer opens the processed version instead of the raw extraction. They validate it, and the workflow picks back up as before.

Here's a real example. OCR reads an invoice total as the text "3,082," pulls "1234" for an AWB number, and comes up empty on the port field. Run that through business logic first, and the reviewer sees something different:

The reviewer is checking business data at that point, not fixing what OCR got wrong.

Enable Human Review also comes with reviewer assignment. Send a document type to a single owner, open it to a group, or round robin it across your team, so work doesn't just default to whoever clicks fastest.

Where this is useful

Freight and logistics operations Shipment records come in with inconsistent AWB numbers, missing ports, and totals formatted as text instead of numbers. Normalize those fields, apply defaults like a home port, and standardize entry types before the record reaches a reviewer. Your reviewer checks whether the shipment is correct, not whether OCR read the manifest properly.

Customs brokers Commercial invoices and entry summaries often arrive with declared values as unformatted text and HS code fields that OCR left blank. Run formatting and lookup logic before review, so the reviewer is validating accuracy and compliance, not fixing number formats under deadline pressure.

Lending and mortgage Income verification sheets and closing cost breakdowns extracted through OCR frequently need dates normalized and fields filled in from defaults. Handle that before the file reaches a reviewer, so their time goes toward checking the numbers against the loan file, not toward reformatting them first.

Manufacturing and procurement Purchase orders and GRNs pulled from supplier documents rarely come out clean. Quantities land as text, supplier codes go missing, and units of measure vary by source. Normalize that first, and your procurement reviewer spends their time confirming the order is right, not correcting how OCR formatted it.

Note: Human Review is now off by default and only fires when you enable it after your processing step. This replaces the previous behavior where it triggered automatically right after OCR.