Global Law Firm Cuts 3M Documents to 440K, Achieving HSR Second Request Compliance in 11 Weeks

What They Needed

On behalf of a healthcare technology company, a law firm needed to review more than 3 million documents in 11 weeks under a Hart-Scott-Rodino (HSR) Second Request. Broad search terms meant more than 50 percent of the data population was potentially privileged.

How Lighthouse Did It

Reducing the Responsive Set and Identifying Privilege

Lighthouse used a proven, court-approved technology-assisted review (TAR) approach to minimize costly and error-prone human review. Lighthouse technology quickly integrated Skype and Microsoft Teams chat data into the TAR workflow.

Subject matter experts from outside counsel coded document sets to train the AI-powered models. Separate models were used to reduce the responsive set and to perform privilege detection. Once the models were trained, they rapidly and accurately analyzed the entire population of documents, achieving a recall rate of 76 percent and a precision rate of 79 percent.

During privilege review, documents on which humans and models disagreed were routed to second-level privilege review. After this quality control step, Lighthouse applied its privilege log and name normalization software, helping outside counsel produce the final privilege log faster.

Identifying Sensitive Data

Lighthouse created regular expressions, which were used in conjunction with AI to find documents containing protected health information (PHI) or personally identifiable information (PII), which were then marked for redaction without manual review. Lighthouse AI technology also identified potentially toxic communications and flagged them for additional review before production.

Results

• 80 percent reduction in the responsive set, eliminating 1.3 million documents
• 51 percent reduction in eyes-on review for PII/PHI
• Eliminated manual review of chat data  
• Identified toxic communications for review before production, typically impossible in Second Request reviews
• 45 percent reduction in privilege review
• Identified 55 percent of privilege QC documents incorrectly coded by human reviewers
• 19,000 documents added to the privilege log efficiently and accurately using Lighthouse technology