AI E-Discovery & Document Review in Legal

Technology-Assisted Review (TAR) uses machine learning to classify documents as relevant, not relevant, or privileged, which reduces the number of documents that require human review. In TAR 2.0 (continuous active learning), the system presents the most informative documents to reviewers first, learns from each coding decision, and continuously re-ranks the remaining documents. This means reviewers see the most relevant material early and the model improves with every decision. A collection of 500,000 documents might require reviewing only 50,000-100,000 to achieve high recall on the relevant population.

Courts have broadly accepted TAR as a valid and often preferred review methodology. Magistrate Judge Andrew Peck's 2012 decision in Da Silva Moore v. Publicis Groupe was the first to approve TAR, and subsequent decisions in Rio Tinto v. Vale and Hyles v. New York City have reinforced its acceptance. Some judges now question the proportionality of insisting on manual review when TAR would be faster, cheaper, and equally accurate. The key requirement is transparency: parties must disclose their TAR methodology, seed sets, and validation approach.

Relativity is the market leader with its aiR for Review (generative AI) and Active Learning capabilities, used by most Am Law 100 firms and major corporations. Everlaw is the leading cloud-native challenger with strong AI features and an interface popular with litigation teams. Reveal offers advanced AI clustering and concept search. Nuix provides data processing and analytics for complex investigations. The market is consolidating: Disco was acquired by CSS in 2023, and established players are rapidly integrating generative AI capabilities for document summarization and review assistance.

Generative AI adds capabilities that TAR alone cannot provide: natural language document summaries that help reviewers understand context faster, automated privilege log drafting that describes the basis for privilege assertions, theme identification across document populations, timeline generation from communication chains, and narrative construction from key documents. Relativity's aiR for Review can explain why a document is relevant in plain language. These capabilities reduce the cognitive load on reviewers and improve consistency, but they supplement rather than replace the core TAR classification workflow.