invoice line discount not extracted: net vs gross amount mismatches

TL;DR

When extraction skips the line-discount column, net amounts compute wrong and the invoice fails the gross-minus-discount-equals-net check. Detect it by validating every line: gross minus discount should equal net within rounding, and the sum of nets should match the invoice total. When the check fails systematically, the discount column is being missed: retrain or reconfigure the extractor to capture it, including discount-as-percentage vs fixed-amount variants. Until the extractor is fixed, flag affected invoices for manual review rather than posting wrong nets.

The query

invoice line discount not extracted: net vs gross amount mismatches

Use this when

• Extracted invoice amounts not reconciling
• Line discounts missing from OCR output
• Net vs gross mismatches on invoice lines

Not for

• Header-level or invoice-level discounts (different handling)
• Tax-inclusive vs exclusive confusion (separate check)

Steps

1. Run the line-level arithmetic check: gross minus discount equals net, per line. Expected output: Failures pinpoint the missed discount column.

2. Confirm the pattern: discounts present on the image but absent in output. Expected output: The extractor is skipping the column, not the data being absent.

3. Reconfigure the extraction template or model to capture the discount field. Expected output: Discounts appear in the output.

4. Handle both percentage and fixed-amount discount presentations. Expected output: Both variants extract correctly.

5. Hold affected invoices for review until the fix is verified on samples. Expected output: No wrong nets post in the meantime.

Variant phrasings

invoice line discount not extracted OCR

net gross mismatch invoice extraction

discount column missing invoice OCR

Root cause

The arithmetic check catches it because gross, discount, and net are a closed triple: if one is missing, the equation breaks. This makes line discounts self-diagnosing, which is why the validation-first approach finds the problem faster than eyeballing output.

Edge cases

• Stacked discounts (line plus header) need both captured; check the invoice total too
• Discounts shown as negative lines rather than columns need a different extraction rule

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_a6sXpC30KwPcPGp8SaeCAA