Scan quality is the single biggest predictor
Native digital PDFs convert reliably with well-preserved structure. Scanned PDFs vary wildly. Documents scanned at 300 DPI or higher with straight alignment produce usable Markdown. Anything below 200 DPI, skewed, or with handwritten annotations will generate output that needs significant manual correction. Before committing to a large migration, audit your source PDFs by type: what percentage are native digital vs scanned, and at what quality levels?
Tables are the hardest problem
Complex tables — especially those with merged cells, nested headers, or multi-page spans — are the most frequent source of conversion failure. Simple grid tables survive well. For complex tables, the best strategy is often to accept imperfect Markdown table output and plan for human review of table-heavy documents. A practical approach: flag documents with more than 5 tables for priority review.
Multi-column layouts need special handling
Academic papers, newsletters, and brochures with multi-column layouts confuse most PDF parsers. The text extraction order often interleaves columns rather than reading column A then column B. MarkItDown Online's layout-aware parser handles common two-column layouts, but three-column and magazine-style layouts still benefit from pre-segmentation or manual review.
The 80/20 rule of batch conversion
In a typical enterprise batch, roughly 80% of documents convert with acceptable quality on the first pass. About 15% need light cleanup (fixing headings, removing repeated headers). About 5% require significant manual reconstruction. Designing your pipeline around this distribution — automated pass for the 80%, quick review for the 15%, and dedicated handling for the 5% — is far more efficient than trying to achieve perfect conversion for every document.