Overview
Procurement teams at large energy companies manage thousands of supplier contracts with significant financial and operational risk buried in dense legal language. This system uses large language models to read, extract, and reason over contract text — dramatically reducing the time analysts spend on contract review.
Problem Statement
Shell India processes 2,000+ procurement contracts annually. Key risk clauses, non-standard terms, and pricing anomalies are frequently missed in manual reviews. The goal: reduce contract review time from 4 hours to 20 minutes per document, while increasing risk detection coverage.
Solution Architecture
A multi-stage pipeline:
- Ingestion: PDF contracts parsed via Azure Document Intelligence
- Extraction: LLM extracts structured fields — parties, values, terms, obligations, penalty clauses
- Risk Scoring: Clause-level risk scoring against Shell's contract standards library
- Benchmarking: Supplier terms compared against historical contract database
- Alerting: High-risk contracts flagged for human review with highlighted clauses
Current Status
In active development. Pilot running with the indirect procurement category.