Law library
eDiscovery: Finding Needles in Legal Haystacks
The average enterprise lawsuit generates 5-15 million documents during discovery. Traditional keyword search returns 70-80% irrelevant results, forcing associates to spend weeks reviewing irrelevant material. Technology-assisted review using continuous active learning has transformed this process. Relativity's AI-powered review platform, deployed at 95% of Am Law 200 firms, reduces document review time by 60-70% while improving recall from 60% to 85%. In a 2025 study of 500 litigation matters, TAR identified 96% of responsive documents using only 20% of the document set for training, compared to 75% recall for exhaustive manual review at 100% of documents.
Case Outcome Prediction
Predictive models trained on historical case data are achieving courtroom-usable accuracy. Lex Machina, now part of LexisNexis, analyzes 200 million+ court docket entries to predict case outcomes with 80-85% accuracy for certain practice areas. Patent infringement cases are particularly predictable, with models achieving 87% accuracy based on judge assignment, patent history, and litigant behavior. A 2026 Stanford study found that lawyers using prediction models settled cases 23% earlier and at 18% lower settlement amounts compared to unaided negotiation, suggesting that data-driven settlement ranges reduce the uncertainty premium.
Automated Brief Writing
Large language models fine-tuned on legal corpora can now generate first-draft briefs, motions, and memoranda. Casetext's CoCounsel, built on GPT-4 and trained on 25 million legal documents, generates draft motions in 5-10 minutes that pass internal quality reviews 85% of the time with minor edits only. The most impactful application has been in immigration law, where AI-drafted briefs have reduced preparation time from 12 hours to 45 minutes per case, enabling pro bono clinics to handle 4x the caseload with the same attorney resources.
Contract Analysis and Due Diligence
Merger due diligence that once required teams of 20-30 attorneys working for 6-8 weeks can now be completed in 7-10 days with AI assistance. Kira Systems processes 10,000 contracts in under 24 hours, extracting 140+ data points per contract including termination rights, assignment clauses, and change-of-control provisions. Error rates average 3-5% versus 8-12% for manual extraction, and the AI improves over time through feedback loops.
Ethical Considerations and Court Acceptance
The 2025 ABA Formal Opinion 512 requires lawyers to maintain reasonable competence in AI tools, including understanding their limitations and verifying outputs. Some courts now require disclosure of AI use in briefs. The Fifth Circuit's 2025 standing order mandates that any filing containing AI-generated text include a certification that a human attorney has verified all citations and legal authorities, a direct response to the 2023 Mata v. Avianca case where ChatGPT hallucinated nonexistent decisions.