Wrong values extracted

Problem: A field has the wrong value (e.g. invoice number is "123" but should be "456").

Solution:

• Edit the value directly in the document view (click the cell, type the correct value, save)
• Check the original document to see what the correct value should be
• Re-process if the document itself was unclear or low quality

Missing values

Problem: A field that should have a value is empty.

Solution:

• Check the document to see if the value is actually there
• Edit and add the value manually if it's in the document
• Improve field description in Fields & Tables to help the AI find it
• Check confidence scores (if enabled) to see if the field had low confidence

Values in wrong fields

Problem: A value appears in the wrong field (e.g. vendor name in invoice number field).

Solution:

• Edit the values to move them to the correct fields
• Improve field names to be more specific (e.g. "Invoice Number" instead of "Number")
• Add AI Descriptions to clarify what each field should contain

Table data mixed up

Problem: Table columns have wrong values or rows are missing.

Solution:

• Edit table cells to correct values
• Check column names match what's in the document
• Use clearer column names (e.g. "Unit Price" instead of "Price")
• See Table Extraction Tips for more help

Use clear field names

Good: "Invoice Number", "Total Amount", "Vendor Name"
Avoid: "Field 1", "Data", "Info"

Clear names help the AI understand what to extract.

Add AI Descriptions

In Fields & Tables, add AI Description text that explains what the field is:

• "The invoice number, usually at the top right"
• "The total amount including tax, usually at the bottom"
• "The vendor or supplier name, usually in the header"

Use the right field types

• Amounts, quantities → Number
• Dates → Date
• Text, names, addresses → Plain text

Using the correct type helps extraction accuracy.

Enable confidence scores

In Parser Settings → Parsing Configuration, enable Calculate confidence scores. This shows which values might be wrong (low confidence = might need review).

Note: This uses 1 credit per page.

Improve document quality

• Use clear, readable scans or PDFs
• Avoid blurry or low-resolution images
• Use consistent document formats (same invoice layout, etc.)

Review and correct

• Review extracted data regularly
• Edit incorrect values
• Re-process documents after improving fields/descriptions