AI E-Discovery & Document Review in Legal

E-discovery — the process of identifying, collecting, processing, reviewing, and producing electronically stored information (ESI) in litigation and investigations — was the first legal function to adopt AI at scale, and it remains one of the most impactful applications. The volume of data involved in modern litigation has grown exponentially: a single custodian may have 100,000+ documents across email, chat, cloud storage, and collaboration platforms. Without AI, reviewing this volume requires armies of contract reviewers at costs that can reach millions of dollars per matter.

Technology-assisted review (TAR) uses machine learning to prioritize and classify documents based on relevance, privilege, and issue coding. TAR 1.0 used a seed set approach where reviewers coded a sample and the model learned from it. TAR 2.0 (continuous active learning) improved on this by learning continuously as reviewers code documents, prioritizing the most informative documents for review and achieving high recall with far fewer documents reviewed. Modern platforms incorporate generative AI to summarize documents, explain relevance decisions, draft privilege logs, and identify key themes across large document populations.

Relativity (now with aiR for Review) and Everlaw are the dominant platforms, with Reveal, Disco (acquired by CSS), and Nuix serving specific market segments. The accuracy of TAR has been validated in numerous court decisions — starting with Da Silva Moore v. Publicis Groupe (2012) and reinforced by Rio Tinto v. Vale (2015) and subsequent rulings that recognized TAR as equal or superior to manual review. Courts have increasingly accepted TAR-based review protocols, and some judges now question parties who insist on manual review of large collections, given the cost and accuracy advantages of AI-assisted approaches.