Best Practices

Improving extraction accuracy

Write clear field descriptions. The description you give a field isn't just documentation for other users - it's also used to help the platform understand what it's looking for. A specific description ("the four-digit cost centre code printed in the top-right box") extracts more reliably than a vague one ("the code").
Use Catalog fields for known values. Whenever a field's value comes from a fixed, known set (statuses, departments, supplier codes), attach a Catalogue instead of leaving it as free text - matching against a known list is far more reliable than open-ended text recognition.
Reserve Required for what truly matters. Every required field with a low-confidence result surfaces during review. Marking too many fields as required creates noisy reviews that make it harder to spot the corrections that actually matter.
Use Field Rules for edge cases, rather than trying to solve everything through the field description alone - rules exist precisely to handle formatting quirks, unit conversions, or organisation-specific conventions.
Test on a small batch first. Before running a few hundred pages through a brand-new Document Type, run a handful through first and check the results. It's much cheaper to fix a Document Type after five documents than after five hundred.

One Document Type per distinct layout, not per department or use case. If two kinds of documents have meaningfully different fields, they should be separate Document Types.
Name things consistently. Clear, consistent Document Type, Catalogue, and Export Preset names make them easier to find as your library grows.
Use Linked Documents to keep related records (like a purchase order and its invoice) connected, rather than relying on matching document numbers manually.

Use Review Mode on the document detail page to jump straight to fields that need your attention rather than checking every field on every document.
Use bulk approve in a batch job for documents flagged No Issues - you don't need to open every single document if the platform is already confident.
Save Search Presets and Export Presets for anything you do more than once - the small setup investment pays off quickly.
Choose Lazy processing for large batches. It's slower, but the accuracy trade-off is usually worth it once you're past small volumes.

	Instant	Lazy
Best for	Small batches (under ~200 files), urgent turnarounds	Large batches (200+ files), when accuracy matters most
Speed	Faster	Slower
Accuracy	Good	Higher