Undisclosed Am Law 100 Firm

A three-attorney team at a leading Am Law 100 firm faced reviewing 126,000 documents for production in a large-scale government investigation with a short timeframe and limited budget. A traditional managed review would have required approximately 20 contract attorneys working for four weeks. Manually coding documents across nearly two dozen different issue codes was described as an arduous task for human reviewers.

The firm deployed EverlawAI Assistant Coding Suggestions, an LLM-powered e-discovery tool, partnering with managed services provider Right Discovery to develop validation workflows. Attorneys crafted and iterated natural-language prompts describing the case context and coding criteria across three stages, then ran the model at scale across the full 126,000-document set once accuracy targets were met.

The team of five (three attorneys plus two support staff) completed the review in approximately one day after prompt iteration, achieving 90%+ accuracy matching or outperforming first-level human reviewers. Review time was reduced by 50–67% compared to traditional methods, and only a quarter of the personnel of a comparable managed review was required. Coding Suggestions also demonstrated greater consistency than human reviewers across identical documents.

• GenAI prompt iteration (roughly 15 hours) is more accessible than crafting complex Boolean search terms, lowering the technical barrier for legal teams adopting AI-assisted review.
• A four-tier classification system (Yes / Soft Yes / Soft No / No) allows attorneys to focus human review effort on ambiguous documents, dramatically improving efficiency.
• Prompts and AI rationales created during initial review serve as a durable record that simplifies future production requests in long-running government investigations.