How to test document intelligence across a commercial loan file
Use the same borrower files and ask the same questions. Vendor pages describe the offer; they don’t show accuracy on your own files.
- Same borrower packet
Run the exact same commercial packet (business tax return, P&L/balance sheet, bank statements, rent roll) across every finalist.
Ask vendors whose reviewed pages highlight mortgage workflows (such as AWS Analyze Lending, Ocrolus, or Hyperscience) to demonstrate commercial financial statements side by side with Docsumo or Instabase.
Sources: Docsumo · Amazon Textract · Ocrolus · Hyperscience · Instabase
- Cross-document conflict flag
Introduce one deliberate mismatch between the borrower application and tax return or bank statement.
Verify whether cross-document checks require higher tiers (Docsumo Enterprise checks against the application and each other; Rossum Ultimate lists multi-document transactions) and ask Google Document AI / Azure Document Intelligence whether built-in checks or custom workflow logic cover the same case. The reviewed pages do not settle that question.
Sources: Docsumo · Rossum · Google Document AI · Azure AI Document Intelligence
- Page & row provenance
Open a flagged discrepancy and trace it to the source page and row.
Confirm every flagged discrepancy links back to the exact source document page and row for underwriter audit (Infrrd states an action/correction audit trail linked to source document page and row); ask other vendors such as Mindee how coordinate or confidence metadata is exported.
Sources: Infrrd · Mindee
- Exception routing & LOS export
Follow an exception from the reviewer back into your loan system.
Trace how a low-confidence or missing document reaches a human reviewer and posts back to Encompass, nCino, or MeridianLink via API/webhooks (Docsumo claims Encompass/MeridianLink/nCino/Blend support; Ocrolus lists LOS integrations).
Sources: Docsumo · Ocrolus
- Written unit-rate quote
Separate OCR from specialized parsers and setup fees.
Google lists OCR at $1.50/1k pages (first 1k free; $0.60/1k above 5M) and bank statements at $0.75/classified document. Azure Q&A gives approximate OCR guidance of ~$1.50/1k pages; region, currency and agreement vary, so confirm in the Azure calculator. Lido Scale is $7,000/year upfront for 42,000 pages/year; Rossum Starter starts at $18,000/year with a one-year minimum. Docsumo adds setup fees.
Sources: Google Document AI · Azure AI Document Intelligence · Lido · Rossum · Docsumo