Fortune 500 industrial company's privilege proof of concept marks 145,000 of 225,000 documents nonprivileged and reports 99.9% accuracy

The general counsel of a Fortune 500 industrial company needed to reduce routine, manual privilege review work so the legal team could focus on higher-judgment questions. Privilege review consumed the most time in their e-discovery projects and carried significant risk—missing even a single privileged document could prove critical to a matter’s outcome.

The GC ran a proof of concept of Text IQ for Privilege in parallel with traditional human privilege review. The AI received the same inputs as contract attorney reviewers: 225,000 documents, a list of 297 known attorneys, the Am Law 200, and legal terms. Relativity describes the tool as using unsupervised machine learning and proprietary socio-linguistic hypergraph technology. Text IQ scored documents from 0 to 1 for indicia of privilege and produced natural-language reason reports explaining why each document was potentially privileged. Relativity's post, first published in 2022 and republished in September 2023, says Relativity acquired Text IQ in 2021 and has since renamed the integrated product Privilege Identify.

Within two and a half weeks, Text IQ identified 80,000 potentially privileged documents and marked 145,000 as nonprivileged, eliminating them from further review. The AI also found 66 attorneys missing from the known attorney list (41 in-house and 25 outside counsel across seven firms), including from first names only, communications between non-attorneys referring to an attorney, and German-language communications. Relativity's post says Text IQ delivered its privilege review in less time than human reviewers and with significant savings, "with documented accuracy of 99.9 percent"; it does not say what the accuracy was measured against. The GC said that otherwise "hundreds and hundreds of documents that were privileged" would have been turned over.

• • Parallel human-vs-AI POCs can build confidence before fully automating high-risk privilege decisions.
• • AI relationship and context modeling can surface privileged custodians and communications humans miss from incomplete attorney lists.
• • Natural-language reason reports explain why each document was flagged as potentially privileged, which the post says helps teams perform QC.