AI Extraction Issues FAQ
Poor Extraction Quality
Section titled “Poor Extraction Quality”Low accuracy: Document quality issues are often the culprit. Ensure resolution meets 150 DPI minimum, text appears clear and readable, the document is properly oriented, and no watermarks obscure content. Request higher-quality scans or try alternative extraction models if problems persist.
Partial extractions: When some fields appear but others don’t, the AI may struggle with unusual layouts or unclear labels. Manually enter missing data, flag the issue for improvement, or verify format compatibility.
Incorrect amounts: “OCR or parsing errors” frequently cause numerical mistakes. Watch for decimal/thousand separator confusion (1.000 vs 1,000) and currency symbol misinterpretation. Compare against the PDF source and correct manually.
Missing Fields
Section titled “Missing Fields”Vendor extraction failures: Vendors may not be recognized if they’re absent from master data, spelled differently, or represented only by logos. Select the vendor manually, add new ones as needed, or create aliases for future matching.
Invoice numbers: Unusual formatting combining letters and numbers, multiple reference numbers, or unexpected placement can confuse extraction. Enter numbers manually or flag for improvement.
Line items: Complex table layouts, multi-page tables, or non-standard structures prevent recognition. Add items manually or consider simpler invoice formats going forward.
Wrong Vendor Matching
Section titled “Wrong Vendor Matching”Similar vendor names or multiple branches cause incorrect selection. Manually choose the correct vendor and create aliases to prevent recurrence. If vendors always show as “Unknown,” verify automatic matching is enabled and the vendor exists in your database.
Currency and Amount Errors
Section titled “Currency and Amount Errors”Currency detection: Ambiguous symbols like $ require manual selection. Set default currencies for specific vendors to streamline processing.
Decimal formatting: Regional differences (European 1.234,56 vs US 1,234.56) cause errors. Verify and correct amounts, then set vendor locale preferences.
Tax calculations: Verify tax rates and whether taxes are included in line prices before editing amounts manually.
Date Format Issues
Section titled “Date Format Issues”Dates swap when formats are ambiguous—01/02/2024 could mean January 2nd or February 1st depending on region. Verify against source documents and set vendor date format preferences.
Re-extraction
Section titled “Re-extraction”Access the re-extract feature when extraction quality improves through model updates or when initial processing had issues. Choose between full document, lines-only, or header-only extraction scopes. “Save your manual edits first (they’ll be overwritten).”
Improving Extraction
Section titled “Improving Extraction”Flag issues directly in the platform by selecting the problem type and adding details. Best results occur with high-resolution PDFs, standard layouts, and machine-generated text. Structured e-invoice formats (PEPPOL, ZUGFeRD) eliminate extraction needs entirely through structured data.