Two-Stage AI Extraction

Every PDF goes through a multi-pass pipeline powered by GPT-4o vision.

1Document Classification

We render your first few pages and identify the document type: Offering Memo, T12, Rent Roll, or Mixed. This tells us exactly which fields to look for.

2Type-Specific Extraction

A targeted prompt extracts 27+ financial fields from up to 10 rendered pages. If key fields are missing, a fallback pass scans the full document for financial tables.

What We Extract

Accuracy by Document Type

Clean digital PDFs (broker-prepared OMs)96%+
Scanned documents and older formats80-90%
Summary-only OMs (limited financials)Partial

Summary-only OMs may be missing line-item expenses or detailed rent rolls. You can fill in any gaps manually.

Confidence Indicators

Every extracted value includes a confidence signal so you know what to trust.

Green — Found directly in the document
Yellow — Inferred or calculated from other values
Gray — Not found; using industry defaults

What To Do When Extraction Is Wrong

Every value in your report is editable. Open the assumptions editor, change any number, and the underwriting model re-runs instantly. No need to re-upload — just correct and go.

Our Test Results

Tested across 10 real offering memos from CBRE, Marcus & Millichap, and local brokerages. Properties ranging from 10 to 79 units across California and Washington markets. The three-pass extraction pipeline achieves 96% field-level accuracy across all 27+ extracted fields.

See for yourself — upload any OM and we'll show you what we find.

Try Free Analysis