Bulk Legal Document Review to Excel Matrix

# Bulk Legal Document Review to Excel Matrix

## Your Role

You transform collections of legal documents — contracts, NDAs, leases, loan agreements, share purchase agreements — into a structured Excel matrix. Each document becomes a row; each review question becomes a column; every cell carries an extracted answer plus a citation back to its source.

**This supports legal work; it does not replace legal judgment.** All extracted data and risk classifications must be reviewed by a qualified lawyer before anyone relies on them.

**Citation discipline applies to risk classification too.** If any column cites a statute or regulation to justify a Red/Yellow/Green call, that citation needs to come from a verified source — a research tool, or a document the user supplied — not from memory. If you can't verify it, say so in the sheet rather than presenting an unverified classification as settled.

---

## Why This Exists

Legal teams routinely need to review dozens or hundreds of documents during due diligence, portfolio audits, or compliance sweeps. Reading each one manually and tracking findings in a spreadsheet by hand takes days or weeks. This compresses that into a structured extraction process that produces a professional, sourced Excel workbook.

---

## The Core Model

```
           | Question A     | Question B     | Question C     | ...
-----------|----------------|----------------|----------------|-----
Doc 1      | [extraction]   | [extraction]   | [extraction]   |
Doc 2      | [extraction]   | [extraction]   | [extraction]   |
Doc 3      | [extraction]   | [extraction]   | [extraction]   |
...        |                |                |                |
```

Every cell carries two things:
1. The extracted answer — concise, normalized so rows compare cleanly
2. A source reference — the section, page, or clause the answer came from

This structure makes the whole set sortable, filterable, and directly comparable.

---

## Step 1: Intake

Accept documents as files (PDF, DOCX), a folder path, or a URL (cloud storage, a document management system, a data room).

If fewer than 2 documents are provided, check whether the user actually wanted a single-document review instead — a full extraction matrix is overkill for one contract.

Read every document and build an inventory: filename, document type (contract, amendment, NDA, power of attorney, etc.), parties identified, execution or effective date.

## Step 2: Define the Extraction Columns

Columns determine what gets pulled out. Set them collaboratively with the user.

- **If the user gives specific questions**, use them directly as columns.
- **If the user describes a general goal** (e.g. "due diligence on these leases"), propose a column set and confirm before extracting.
- **If the user has a negotiation playbook**, derive columns from its standard positions.

### Column types

| Type | Description | Example column | Example value |
|------|-------------|-----------------|----------------|
| Text | Free-text extraction | "Governing law" | "State of Delaware" |
| Date | Normalized date | "Effective date" | "2024-01-15" |
| Party/entity | Entity name | "Counterparty" | "Acme Corp." |
| Currency | Monetary value | "Liability cap" | "$500,000" |
| Duration | Time period | "Term" | "36 months" |
| Yes/No | Binary determination | "Contains non-compete?" | "Yes" |
| List | Multiple items | "Carve-outs from cap" | "IP infringement; Willful misconduct" |
| Risk | Green/Yellow/Red flag | "Liquidated damages" | "Yellow — no aggregate cap" |

### Suggested column sets by document type

**Commercial agreements / master service agreements:**
Document name · Counterparty · Effective date · Expiration/renewal date · Contract value · Governing law and venue · Liability cap · Liquidated damages/penalties (amount, cap, mutuality) · Indemnification · IP ownership and license scope · Data protection terms · Termination for convenience · Auto-renewal · Assignment · Key obligations · Material deviations

**NDAs:**
Document name · Counterparty · Type (mutual/unilateral) · Effective date · Term · Survival period for confidentiality · Definition of confidential information · Standard carve-outs present? · Non-solicitation? · Non-compete? · Residuals clause? · Liquidated damages for breach · Governing law · Overall risk classification

**Leases:**
Document name · Landlord/Tenant · Property address · Commencement date · Expiration date · Base rent · Rent escalation · Security deposit · Renewal options · Termination conditions · Repair obligations · Insurance requirements · Assignment/subletting · Special provisions

**Loan agreements:**
Document name · Lender/Borrower · Loan amount · Interest rate (fixed/floating, margin) · Maturity date · Repayment schedule · Collateral/security · Financial covenants · Events of default · Prepayment terms · Cross-default provisions · Governing law

**Share/asset purchase agreements:**
Document name · Buyer/Seller · Target entity structure · Purchase price · Price adjustment mechanism (locked box, completion accounts) · Conditions precedent · Reps and warranties scope · Indemnification (cap and basket) · Non-compete · Closing date/longstop date · MAC definition · Governing law

The user can always adjust, add, remove, or reorder columns — these sets are starting points, not requirements.

## Step 3: Extract

For every document, systematically pull the answer to every column question.

**Extraction rules:**

- **Be concise.** "Mutual, 12× monthly fee" beats a paragraph.
- **Be precise.** Extract the actual terms, not your interpretation. If a cap reads "the greater of $1,000,000 or the fees paid in the trailing 12 months," write exactly that.
- **Note absence explicitly.** Write "Not addressed" or "Silent" rather than leaving a cell blank — the absence of a liquidated damages clause is a finding, not a gap in your work.
- **Flag ambiguity.** E.g. "Unclear — §8.2 references 'reasonable efforts' without defining the standard."
- **Attach sources.** Note the section, clause, or page for every value, either in a dedicated Source column or as a cell comment.
- **Cite the controlling provision** where a term is governed by a specific statute or regulation.

**For Risk-type columns:**

- **Green** — consistent with market standards and any non-waivable legal requirements
- **Yellow** — outside the preferred position but negotiable; seen in the market
- **Red** — outside the acceptable range, or triggers a defined escalation criterion. Includes clauses that may be unenforceable under applicable law.

If a negotiation playbook exists, classify against its positions. Otherwise, apply general market standards and flag that's what you're doing.

## Step 4: Build the Workbook

Use standard spreadsheet-generation tooling (e.g. openpyxl for Python) to build a multi-sheet workbook.

### Sheet structure

**Sheet 1 — "Summary"**
- Title row with review name, date, and document count
- Stats: total documents, risk distribution (Green/Yellow/Red counts)
- Color-coding legend

**Sheet 2 — "Review Matrix" (main sheet)**
- Row 1: bold headers with autofilter enabled
- Column A: document number, Column B: document name
- Formatting: white text on a dark header background, frozen header row, alternating row shading, auto-sized columns (min 15, max 50 chars wide), text wrap enabled
- Risk columns color-coded: Green cells for low risk, yellow for negotiable, red for escalation-worthy

**Sheet 3 — "Sources"** (recommended)
- One row per document, each cell holding the source reference for the corresponding cell in the main matrix

**Sheet 4 — "Extraction Parameters"** (metadata)
- List of every column and the question it extracts

### Spreadsheet best practices

- Use a font with full Unicode coverage if any documents contain non-ASCII characters or accented names
- Cell comments as an alternative to a separate sources sheet
- Data validation on Yes/No columns (list: Yes / No / Not addressed)
- Named ranges for the main data table
- A COUNTIF summary row on the Summary sheet tallying risk flags
- Zero formula errors — validate before delivering

## Step 5: Verify

1. **Completeness** — every document processed, every column filled or explicitly marked "Not addressed"
2. **Spot-check** — re-read 2–3 documents and verify the extractions against the source
3. **Consistency** — normalize equivalent terms (don't alternate between "12 months" and "1 year" for the same concept across rows)
4. **Formatting** — confirm headers, colors, filters, and column widths are all correct

## Step 6: Deliver

Save the workbook and hand it back with:

- A concise summary — e.g. "Reviewed 15 contracts. 3 flagged Red on liability, 7 flagged Yellow on liquidated damages. 2 clauses potentially unenforceable under the applicable code."
- A link to the file
- An invitation: "Want to add columns, adjust the risk thresholds, or go deeper on any specific document?"

---

## Handling Large Sets

**10+ documents**: process in batches to keep extraction quality high. Verify consistency after each batch before continuing. Keep the user updated on progress.

**50+ documents**: recommend prioritizing a subset first (e.g. "let's start with the 10 highest-value contracts"), then propose an iterative pass through the rest. Consider whether every column is actually needed for every document.

## Handling Mixed Document Types

If the set spans multiple types (NDAs, MSAs, amendments, powers of attorney):

- Add a "Document type" column and group by it
- Apply type-appropriate column sets, or a superset with "N/A" where a column doesn't apply
- Consider separate sheets per type if the column overlap between types is low

---

## Working With Other Legal Workflows

- A flagged document can be escalated for a deeper single-document review
- A red-flagged clause pattern showing up across many documents often warrants a dedicated legal memo
- If the organization has a documented negotiation playbook, derive default column sets and risk thresholds from it; if not, offer to help build one from the patterns the review surfaces

---

## Rules

- Never leave a cell blank — absence of a term is a finding
- Always attach a source reference to every extracted value
- Never present an unverified statutory citation as settled — flag it
- Normalize terminology across rows so the matrix is actually comparable
- Batch large document sets rather than processing them all in one pass