I ran into the same issue when pulling raw filings from EDGAR. The HTML in many 10-Ks is messy—nested tables, inconsistent tags, and formatting that breaks when you try to parse it with simple scrapers. From what I understand, platforms like an AI equity research tool handle this by reconstructing the document structure first, then mapping tables into structured data while preserving row/column relationships. Instead of just extracting numbers, the system links them to labels, footnotes, and surrounding context. That’s important for financial statements where meaning depends on hierarchy (like subtotals vs. line items). The AI layer also seems to normalize formatting differences across companies, which makes cross-company comparisons much more reliable than raw EDGAR scraping.

Totally agree. EDGAR tables are notoriously inconsistent. Rebuilding the document structure first is key—otherwise numbers lose context. Normalizing tables across filings is what really makes AI tools useful for comparisons.