E-discovery (electronic discovery) uses AI to find relevant documents in litigation — searching, reviewing, and producing electronically stored information (ESI) including emails, documents, chat messages, databases, and social media using machine learning to identify relevant materials, dramatically reducing the cost and time of document review.

What Is E-Discovery?

• Definition: Process of identifying, collecting, and producing ESI for legal matters.
• Scope: Emails, documents, spreadsheets, presentations, chat/messaging, social media, databases, cloud storage, mobile data.
• Stages: Identification → Preservation → Collection → Processing → Review → Analysis → Production.
• Goal: Find all relevant, responsive documents while minimizing cost and time.

Why AI for E-Discovery?

• Volume: Large cases involve millions to billions of documents.
• Cost: Document review is 60-80% of total litigation costs.
• Time: Manual review of 1M documents requires 100+ reviewer-months.
• Accuracy: AI-assisted review is as accurate or more accurate than human review.
• Proportionality: Courts require proportional discovery efforts.
• Defensibility: AI-assisted review is widely accepted by courts.

Technology-Assisted Review (TAR)

TAR 1.0 (Simple Active Learning):

• Senior attorney reviews seed set of documents.
• ML model trains on seed set, predicts relevance for remaining.
• Human reviews AI predictions, provides feedback.
• Iterative training until model stabilizes.

TAR 2.0 (Continuous Active Learning / CAL):

• Start with any documents, no seed set required.
• AI continuously learns from every document reviewed.
• Prioritize most informative documents for human review.
• More efficient — achieves high recall with fewer reviews.
• Standard: Most widely used approach today.

TAR 3.0 (Generative AI):

• LLMs understand document context and legal relevance.
• Zero-shot or few-shot relevance determination.
• Generate explanations for relevance decisions.
• Emerging approach, not yet widely accepted by courts.

Key AI Capabilities

Relevance Classification:

• Classify documents as relevant/not relevant to legal issues.
• Multi-issue coding (relevant to which specific issues).
• Privilege classification (attorney-client, work product).
• Confidentiality designation (public, confidential, highly confidential).

Concept Clustering:

• Group similar documents for efficient batch review.
• Identify document themes and topics.
• Near-duplicate detection for related document families.

Email Threading:

• Reconstruct email conversations from individual messages.
• Identify inclusive emails (final in thread, contains all prior).
• Reduce review volume by eliminating redundant messages.

Entity Extraction:

• Identify people, organizations, locations, dates in documents.
• Map communication patterns and relationships.
• Timeline construction for key events.

Sentiment & Tone Analysis:

• Identify concerning language (threats, admissions, consciousness of guilt).
• Flag potentially privileged communications.
• Detect code words or euphemisms.

EDRM Reference Model

1. Information Governance: Proactive data management policies. 2. Identification: Locate potentially relevant ESI. 3. Preservation: Legal hold to prevent spoliation. 4. Collection: Forensically sound gathering of ESI. 5. Processing: Reduce volume (deduplication, filtering, extraction). 6. Review: Examine documents for relevance, privilege, confidentiality. 7. Analysis: Evaluate patterns, timelines, key documents. 8. Production: Produce responsive documents to opposing party. 9. Presentation: Present evidence at deposition, hearing, trial.

Metrics & Defensibility

• Recall: % of truly relevant documents found (target: 70-80%+).
• Precision: % of documents marked relevant that actually are.
• F1 Score: Harmonic mean of precision and recall.
• Elusion Rate: % of relevant documents in discarded (not-reviewed) set.
• Court Acceptance: Da Silva Moore (2012), Rio Tinto (2015) endorsed TAR.

Tools & Platforms

• E-Discovery: Relativity, Nuix, Everlaw, Disco, Logikcull.
• TAR: Brainspace (Relativity), Reveal, Equivio (Microsoft).
• Processing: Nuix, dtSearch, IPRO for data processing.
• Cloud: Relativity RelativityOne, Everlaw (cloud-native).

E-discovery with AI is indispensable for modern litigation — technology-assisted review enables legal teams to process millions of documents efficiently and defensibly, finding the relevant evidence while dramatically reducing the cost that makes justice accessible.