Skip to main content
Finance & LegalFinancial Advisory37 lines

Financial Data Room Document Extraction

Extract financial tables with page-level provenance, consistent units and reconciliation across source documents.

Quick Summary7 lines
Use this skill to turn authorized data-room documents into a reviewable financial dataset. Establish the transaction purpose, access permissions, entities, periods and requested metrics. Do not upload confidential files to an unapproved extraction service or imply that extraction constitutes an audit.

## Key Points

- [SEC EDGAR data APIs and XBRL](https://www.sec.gov/search-filings/edgar-application-programming-interfaces)
skilldb get financial-advisory-skills/financial-data-room-document-extractionFull skill: 37 lines
Paste into your CLAUDE.md or agent config

Financial Data Room Document Extraction

Use this skill to turn authorized data-room documents into a reviewable financial dataset. Establish the transaction purpose, access permissions, entities, periods and requested metrics. Do not upload confidential files to an unapproved extraction service or imply that extraction constitutes an audit.

Inventory and preserve sources

Preserve the original document and record a stable source ID, filename, version, checksum and received date. Classify financial statements, management accounts, forecasts, contracts and supporting schedules separately. Distinguish audited, reviewed and unaudited materials using the documents' own statements.

Identify the reporting entity, consolidation scope, currency, period, accounting basis and scale on every relevant table. A column labeled FY may not represent the same dates across companies. Keep restatements and superseded versions linked rather than mixing them.

Extract with provenance

Prefer reliable structured source data where available, but verify its context. XBRL facts, for example, require attention to concept, units, period and filing context; a familiar tag alone is insufficient. For PDFs or images, retain page and table references and inspect OCR-sensitive characters.

Store original text alongside normalized values. Handle parentheses, minus signs, decimal separators, footnote markers and thousands or millions explicitly. Missing is not zero. Preserve estimates, ranges and not-applicable entries as distinct states.

Attach each material value to its source page or cell, original unit, transformation and confidence or review status. Keep formulas reproducible. Do not infer a missing segment allocation simply to complete a table.

Reconcile before interpreting

Reconcile before interpreting trends or ratios. Check column totals, subtotals, opening-to-closing movements and appropriate cross-statement relationships. Investigate discrepancies rather than inserting a balancing figure. Timing, scope, currency translation or an excluded component may explain a difference.

Review a sample of straightforward cells and every high-impact or low-confidence extraction directly against the source. Increase review when scan quality or formatting varies. A high OCR confidence score is not proof of financial correctness.

Produce an exception register with amount, source, suspected cause and owner. Separate extracted facts from analytical adjustments and management explanations. Record assumptions used in any derived metric.

Deliver the normalized workbook or dataset, field dictionary, provenance index and reconciliation results. Clearly identify unverified documents and unresolved differences so downstream users cannot mistake a complete-looking table for fully validated financial evidence.

References

Install this skill directly: skilldb add financial-advisory-skills

Get CLI access →