Discovery packet intake cuts paralegal time per matter to 75 minutes

Each new matter arrived as a packet — medical records, police reports, prior pleadings, intake form, photo evidence. The intake paralegal spent ~4 hours per matter classifying the packet, extracting key facts, and creating the Clio matter file with iManage folder structure.

What we built

• Document classifier within an intake packet: medical record / police report / pleading / intake form / photo / correspondence.
• Per-class blueprints extracting structured key facts — incident date, location, parties, injury type, dates of treatment, statute of limitations clock.
• Clio Manage matter creation with party/conflict pre-population.
• iManage folder structure auto-generated per the firm’s standard taxonomy.
• Explicit boundary: the agent never drafts substantive legal analysis, never cites cases, never suggests claims. It indexes and routes — paralegals and attorneys do the analysis.

Why this engagement was deliberately narrow

The legal AI hallucination wave through 2025–2026 (Sullivan & Cromwell, Gordon Rees twice sanctioned, 1,300+ documented incidents per Damien Charlotin) drove our design choice to make this a document-processing agent, not a research or drafting agent. The agent never produces text that goes into a court filing. The boundary is contractual, not just technical.

Stack. AWS Bedrock Data Automation · Amazon Textract for scanned court filings · Bedrock single-model for auditability · Clio Manage + iManage REST · S3 with KMS encryption for archive.

Paralegal time per intake fell from ~4 hours to ~75 minutes. 100% of statute-of-limitations dates correctly captured on motor-vehicle and personal-injury intakes. Each paralegal absorbed ~30 additional matters/month without adding hours.