Unique Priv Log Descriptions Drafted by AI Sped Up Privilege Review
Total Key Docs Delivered for Early Case Analysis and Depo Prep
Integrated the latest in analytics and review for faster, more consistent results.
Used regulator-approved, battle-tested AI models for responsive and privilege review tasks.
3.6M starting documents reduced to 670K produced docs.
2.2K unique privilege log descriptions drafted by AI and approved/edited by attorneys.
Document production and privilege logging completed ahead of schedule.
300 key docs (0.05% of produced documents) delivered for early case analysis and deposition preparation.
Two Months to Tackle Three Million Documents
A financial institution responding to a Hart-Scott-Rodino Second Request by the Department of Justice had two months to review 3.6M documents (2.4TB of data) to reach substantial completion.
With that deadline, any time that reviewers spent on irrelevant documents or unnecessary tasks risked pushing them off schedule. So outside counsel called on Lighthouse to help efficiently produce documents to regulators.
AI and Experience Prove Up to the Challenge
Using our Second Request review solution, we devised an approach that coordinated key data reduction tactics, modern AI, and search expertise at different stages of review.
Junk Removal and Deduplication Set the Stage
We started by organizing the dataset with email and chat threading and removing 137K junk documents. Then we shrank the dataset further with our proprietary deduplication tool, which ensures all coding and redactions applied to one document automatically propagate to its duplicates.
AI Model Removes 1.5 Million Nonresponsive Documents
To build the responsive set, we used our AI algorithm, built with large language models for sophisticated text analysis. We trained the model on a subset of documents then applied it to all 2.2M TAR-eligible documents, including transcripts from chat platforms.
The model identified 80% of the documents containing responsive information (recall) with 73% accuracy (precision). The final responsive set consisted of 650K family-inclusive documents—18% of the 3.6M starting corpus.
AI Supports Privilege Detection, QC, and Descriptions
Our AI Privilege Review solution supported reviewers in multiple ways.
First, we used a predictive algorithm in conjunction with privilege search terms to identify and prioritize potentially privileged documents for review.
During QC, we compared attorney coding decisions with the algorithm’s assessment and forwarded any discrepancies to outside counsel for final privilege calls.
For documents coded as privileged, we used a proprietary generative AI model to draft 2.2K unique descriptions and a privilege log legend. After reviewing these, attorneys left nearly 1K descriptions unchanged and performed only light edits on the rest.
Search Experts Surface the 300 Documents Most Important for Case Prep
Alongside the production requirements for the Second Request, Lighthouse also supported the institution’s case strategy efforts. Each tranche of work was completed in 4 days and within an efficient budget requested by counsel, who was blown away by the team’s speed and accuracy. Using advanced search techniques and knowledge of legal linguistics, our experts delivered:
130 documents containing key facts and issues from the broader dataset, for early case analysis.
170 documents to prepare an executive for an upcoming deposition.
Beating the Clock Without Sacrificing Cost or Quality
With Lighthouse Review—including the strategic use of state-of-the-art AI analytics—outside counsel completed production and privilege logging ahead of schedule. The financial institution met a tough deadline while controlling costs and achieving extraordinary accuracy at every stage.