Product · Document extraction
AI extraction that shows its work.
Offering memorandums, rent rolls, and T-12s in — a reviewed, page-cited starting model out. Every value is traceable to its source and nothing applies without your Confirm.
01
Upload the package
Drop in the OM PDF, the rent roll, and the T-12 — spreadsheet or PDF. Spreadsheets parse deterministically; AI is used only where the document is unstructured.
02
Extraction, with receipts
Key OM values come back cited to the page they were found on, with a verbatim quote. PDF statement lines carry a per-line page badge. Anything the model can’t cite confidently is omitted, not guessed.
03
You review, then Confirm
Everything lands in a mapping preview — reconciled against the statement’s own totals — and nothing enters the model until you press Confirm. The AI drafts; the underwriter decides.
What gets extracted
Offering memorandum
Purchase price, unit count and mix, debt terms, and the in-place operating lines — each key value cited to its page with a verbatim quote. New to the document? Read the underwriting guide.
Rent roll
Unit-level detail — units, rents, and status — mapped into the model’s rent roll with automatic column detection you can override before anything applies. New to the document? Read how to read a rent roll.
T-12 operating statement
Revenue and expense lines mapped to the model’s chart, reconciled against the statement’s totals, with per-line page provenance on the PDF path. New to the document? Read how to read a T-12.
Extraction you can audit, not trust blindly
Most AI extraction demos end at “look, it filled in the numbers.” The problem is the number it got wrong — because in underwriting, one wrong operating line moves valuation by six figures. That's why Nivora's extraction is built around provenance: key values carry the page they came from and a verbatim quote from that page, so checking a number takes seconds, not a re-read of the deck.
The same posture runs through the whole import path. A value the model can't cite confidently is omitted rather than guessed — a wrong page is worse than none. Statement imports are reconciled against the document's own totals before you confirm, so a missed line surfaces as a visible discrepancy instead of a silent hole in the model. And OM-internal consistency checks flag where the deck contradicts itself — a unit mix that can't support the stated rent, lines that don't add to the printed totals — before you rely on it.
Where AI isn't needed, it isn't used: spreadsheet rent rolls and T-12s parse deterministically, the same way every time. The result in every case is a reviewed starting model — pro forma, debt, and returns — that an underwriter has confirmed line by line, with an audit trail behind it.
Does the AI write numbers straight into my model?
No. Every extraction lands in a mapping preview where each value is shown with its source — down to a per-line page badge for PDF statements — and nothing enters the model until you press Confirm. The AI drafts; you decide.
What documents can Nivora extract?
Offering memorandum PDFs (purchase price, unit count, debt terms, and the in-place operating lines), plus rent rolls and T-12s. Spreadsheets are parsed deterministically — no AI involved — and PDF statements go through AI extraction with per-line page provenance.
How do I know an extracted number is right?
Key OM values carry a citation: the page number plus a short verbatim quote from that page, so you can check the source in seconds. Values the extraction cannot cite confidently are omitted rather than guessed. Statement imports are additionally reconciled against the document’s own totals before you confirm.
Is my deal data used to train AI models?
No. Documents are processed under API terms that do not permit the provider to train models on your data, and Nivora does not train AI models on your documents either. Extracted values are always shown to you for review. See the privacy policy for the full data-handling statement.
Extraction accuracy varies with document quality; that is exactly why every value is shown for review with its source before it applies. See pricing for plans — extraction is included, not an add-on. Not investment advice.
Extraction feeds the same engine you see on the sample deal: documents to reviewed inputs to a full pro forma with IRR, DSCR, and the LP/GP waterfall. New to the metric? Read how to read a T-12. Not investment advice.
See it on a real deal.
In Nivora this runs off your actual rent roll and T-12 — the full monthly model, reconciled to the penny. See it live on a fictional sample deal.