Results Are In: 5 Lessons From An Independent Study Of aiR For Review

Title

Results Are In: 5 Lessons From An Independent Study Of aiR For Review

Content

An independent head-to-head study by Redgrave LLP tested Relativity aiR for Review against a traditional first-pass managed review workflow on approximately 45,000 documents from a real-world pharmaceutical marketing, controlled substances, and federal compliance data set. The aiR for Review workflow completed the matter in roughly 18 hours of attorney time compared with 1,123 hours from a 24-person team over seven business days, achieving 88 percent recall versus 64 percent for the active learning workflow and a 1 percent elusion rate versus 3 percent. The study's author, writing in a Conventus Law summary of the Redgrave LLP report, said the result matters because faster review is valuable, but faster review that also improves completeness is far more valuable, with aiR for Review delivering both higher recall and lower elusion in the same run. Document review in legal e-discovery has long been bottlenecked by the cost of attorney time on high-stakes, judgment-heavy matters where responsive documents are not simply those that mention a topic but those that contain evidence of compliance with, violation of, or reckless disregard for federal requirements. Relativity aiR for Review is a generative AI review workflow that became generally available in 2024 and was made a standard component of RelativityOne in October 2025, with the Conventus Law summary of the Redgrave study published on June 11, 2026. The platform combines large language model-based reasoning with citation-grounded rationales for each responsiveness prediction, allowing reviewers to inspect AI rationale, score individual documents, and re-rank or override AI calls. In the human-plus-AI follow-up, the subject-matter expert re-reviewed 151 documents where he and aiR for Review disagreed and changed 10 calls to responsive, meaning approximately one in every 15 disagreement documents surfaced something the expert agreed was responsive after further review. Looking ahead, the study frames aiR for Review as a second lens on evidence that supports quality control on close calls rather than a replacement for attorney judgment, with implications for how generative AI slots into high-stakes legal review workflows going forward.