arrow_back Back to Legal Legal Ops Automation

Automated NDA Triage

The Context / Challenge

Setting the scene

Every legal team knows the pattern: an NDA lands in the queue, a team member reads it, classifies it, drafts redlines, and posts the analysis back to the ticket. For standard mutual NDAs with boilerplate terms, this is repetitive work that consumes time without requiring real legal judgment.

The costs are turnaround delays, inconsistent classification across reviewers, and context-switching overhead. The team spends more time formulating redlines than reviewing them.

Our Approach

AI-powered triage with a traffic-light system

The workflow runs on a schedule (every 30 minutes), picks up new NDA tickets, and processes each one through a classification and redline pipeline.

1

Intake

Query the ticketing system for un-triaged NDAs. Download the attachment, convert to .docx if needed, and extract text.

2

Classification

An LLM reads the NDA alongside your team's playbook and classifies it using a traffic-light system.

3

Redlines

For YELLOW/RED, a second LLM pass generates clause-specific redlines with exact quotes, replacement language, and playbook citations. Applied as tracked changes to the .docx.

4

Post and mark

Classification, summary, and redlined doc are posted as an internal comment. The ticket is labeled to prevent reprocessing.

Classification system

GREEN Standard terms

NDA matches your standard template. No redlines needed. Ready to sign.

YELLOW Minor deviations

Contains non-standard clauses that need redlines. AI generates tracked changes for review.

RED Requires attention

Significant deviations or non-standard structure. Flagged for manual review with detailed analysis.

Key takeaways

How to replicate this

01
The playbook is everything. Invest in documenting your review standards before building anything. The LLM defers to the playbook for all substantive rules.
02
Keep prompts mechanical, not substantive. All legal judgment lives in the playbook. Prompts enforce verbatim quoting, no hallucination, and structured JSON output.
03
Start with classification only. Add redlines once GREEN/YELLOW/RED accuracy is solid.
04
Lawyers stay in the loop. The system does prep work, not decision-making. The team spends time reviewing redlines rather than formulating them.
Results

What we achieved

Eliminated reviewing and redlining all new NDAs on the legal service desk. Reduced review time significantly. The team now spends more time reviewing redlines rather than formulating them. Classification consistency improved across the team, and turnaround time on standard NDAs dropped substantially.

Watch the demo

Video walkthrough

Watch video