Roche cuts patent GenAI project cycles from months to days with Themis PatAI

Roche’s patent team in Basel faced a rising volume of European Patent Office case law and needed faster responses to oppositions and appeals. Patent research remained an exhaustive, manual legal process spanning more than 50,000 EPO cases, with new AI tooling blocked by months-long traditional development cycles and IT bottlenecks.

Roche operationalised Agentic AI through Themis PatAI, a unified interface built on Dataiku that routes natural language queries to sub-agents for semantic search, full-text analysis, and internal document lookups via Google Drive. Using Dataiku’s citizen-development approach, non-technical patent attorneys and paralegals can build and deploy complex AI search flows in days without waiting on traditional IT delivery.

New GenAI projects now take days instead of months. Roche reports US$100,000–$250,000 in annual attorney hours saved per case, plus US$375,000–$475,000 saved by keeping development in-house rather than relying on consultancy. The programme currently supports about 80 European attorneys, with planned expansion toward a larger global user base.

• • Empowering subject-matter experts as citizen developers on a governed low-code platform can bypass IT waitlists and accelerate legal AI delivery.
• • Agentic AI that orchestrates semantic search, full-text analysis, and internal document lookup can turn large patent case corpora into a usable research workflow.
• • Keeping AI build capacity in-house can produce material consultancy-cost avoidance alongside attorney-time savings.