A 12-person law firm in Bucharest received 30–40 new client intake documents per week via email — contracts, ID copies, power of attorney forms, case briefs. Each document had to be: read manually → categorized → routed to the right attorney → logged in their case management system → a response sent to the client.
Average time per document: 14 minutes of paralegal time. With 35 documents/week: 490 minutes (8+ hours) weekly on intake routing alone.
The firm had tried hiring an additional paralegal. It helped for 3 months, then volume grew again.
The ArchitectureWe built a Document Intelligence Pipeline:
→ All incoming emails to a dedicated intake address are intercepted by the pipeline
→ Vision AI (OCR layer) reads and classifies each attachment by document type
→ A routing engine matches document type + client name to the correct attorney's queue
→ The case management system is updated automatically with document metadata
→ Client receives an automated acknowledgment within 90 seconds of sending
For 94% of documents, the entire process is zero-touch. The remaining 6% — ambiguous or novel document types — are flagged for 30-second human review with the classification pre-filled.
Deployment timeline: 18 days (including custom integration with their legacy case management system).
The Numbers (60 days post-deployment)The OCR classification model was trained on 200 sample documents from the firm's own archive. It knows their document taxonomy — not a generic model. This is the difference between a tool and a Custom Engine.