PARSER GUIDES

How to Increase Extraction Accuracy

Set up your parser for more consistent extraction results

Use this checklist before processing a large batch. Make one change at a time and test it with a few representative documents.

Use the closest document type

Choose the Document Processing Type that matches the files you receive. Keep different document types in separate parsers.

If one inbox contains mixed files, use a Document Classifier to route them.

Use clear field names

Name fields after the value you want:

• Use Invoice Number, not Number
• Use Total Including Tax, not Total
• Use Supplier Name, not Name

Use a Table for repeating rows such as invoice line items or bank transactions.

Describe fields that could be misunderstood

Add an AI Description when the field name is not enough. Describe what to capture and how to distinguish it from similar values.

Example: "The final amount due after tax and discounts, usually labelled Amount Due."

See Add and Edit Fields for field setup.

Add reasoning only for difficult decisions

For a field that requires a comparison, fallback rule, or calculation, turn on Enable advanced reasoning and add a short instruction under Reasoning logic (optional).

Example: "Choose Balance Due when it is present. Otherwise use Total. Do not use Subtotal."

Do not add reasoning to every field. Straightforward values usually work better with a clear field name and AI Description.

Test representative documents

Test documents from different suppliers, clients, or layouts--not several copies of the same format.

Start with two or three files. Check important fields and table rows before processing the full batch.

Use readable source files

For scans and photos:

• Make sure the text is sharp and readable
• Keep pages straight and fully visible
• Avoid shadows, handwriting over printed values, and very low-resolution images

If the value is difficult for a person to read, extraction may also be unreliable.

Choose processing settings only when needed

If your account shows Data extraction mode, start with Accurate. Use Critical selectively for documents where mistakes are especially costly.

For long or difficult tables, review Advanced Table Extraction.

These options can use additional credits. Check Credits and Billing before enabling higher-cost settings.

Add a review step for important data

Accuracy settings reduce mistakes, but important amounts, dates, account numbers, and approval decisions may still need review.

Use Reviews & Approvals or enable Confidence Scores to help reviewers focus on uncertain values.

If results are already wrong

Use Troubleshoot Extraction Accuracy to diagnose missing values, incorrect fields, or table problems.