Orrick

Orrick needed to evaluate whether generative AI could deliver nuanced, consistent document-review determinations comparable to human review and predictive coding on a live IP matter. Attorney Cal Yeaman wanted evidence of how GenAI would fit into existing ediscovery workflows before relying on it at scale.

Orrick tested and deployed EverlawAI Assistant Coding Suggestions on approximately 10,000 documents in a real-world IP case. The team refined prompts on a small sample first, then used a hybrid workflow combining Coding Suggestions with Predictive Coding—GenAI to speed sampling and predictive coding to validate GenAI accuracy and prioritize human QC.

Coding Suggestions were more accurate than human review by a statistically significant margin; of GenAI not-relevant calls, human reviewers reversed only a single document. Yeaman estimated the AI-powered review process reduced document review costs by more than 50%. The tool also helped the team identify key documents faster and free attorney time for case strategy.

• • Pilot Coding Suggestions on a small sample and iterate prompts before applying them to the full document set.
• • Pair GenAI coding with predictive coding so each method validates and complements the other.
• • Match the tool to the matter: more linear or complex subject-matter analysis can favor LLMs over human-only review.