Reed Smith

Privilege review and privilege logging represent one of the highest costs in e-discovery, and the challenge has persisted despite two decades of process improvement elsewhere in litigation support. Reed Smith faced a massive case involving tens of millions of documents requiring review under significant time pressure, with strict deadlines to produce documents while ensuring privileged materials were not inadvertently disclosed. The stakes were high: a missed privileged document in a production set carries serious professional and legal consequences, making speed-accuracy tradeoffs especially painful for large law firms handling complex, document-intensive matters.

Reed Smith deployed a Technology-Assisted Review (TAR) model trained on a seed set of known privileged documents to score the entire document population for likelihood of privilege. Documents scoring above a defined threshold were routed directly into privilege review queues, bypassing earlier-stage review tiers and eliminating unnecessary multi-level handling. Reviewers working the remaining population could see each document's privilege score in real time, allowing them to calibrate attention without disrupting their workflow. The same model was applied as a quality control layer on production sets — flagging any documents with elevated privilege scores that may have slipped through manual review. The firm also discussed integrating generative AI to provide rationale-based privilege determinations alongside scores, pointing toward future automated privilege logging.

The TAR-assisted workflow made privilege review materially more efficient across a case involving hundreds of thousands of privileged documents within a corpus of tens of millions. Time invested in training and validating the model was recouped on the back end through faster document throughput and reduced redundant review cycles. Key outcomes included:

• Faster production cycles — high-confidence privileged documents moved directly to specialized queues, eliminating unnecessary intermediate review stages
• Improved QC coverage — automated scoring of production sets caught privileged documents that manual review and keyword searches missed
• Reviewer focus — real-time privilege scores directed human attention to documents most likely to require careful analysis, improving accuracy without slowing throughput

• TAR for privilege is low-risk and underused — the tool augments human judgment rather than replacing it; workflow and QC applications carry minimal legal risk and deliver measurable efficiency gains.
• Tiered routing based on privilege scores reduces unnecessary multi-level review, concentrating expert attention where it matters most.
• QC scanning of production sets using the privilege model is a high-value safeguard that catches what keyword searches and manual review miss.
• GenAI's rationale output — explaining why a document is likely privileged — is a meaningful evolution that could reduce training effort and eventually support automated privilege logging.
• Early interrogation of data using GenAI can surface unknown privileged topics before review begins, improving seed set quality and model precision.