How One Reinsurer Cut Risk Analysis Weeks to Hours
A reinsurance risk team wired a hybrid RAG and custom legal dictionary across 10,000 contracts — collapsing 2–3 weeks of analyst work to under a day at 97% accuracy.
2–3 weeks → <1 day
Cut from contract risk analysis
Mike Mayes
Managing Director
Egoless Consulting
Connect ↓
The Challenge
A large specialty reinsurance firm managed 10,000 complex contracts — each worth $100M–$1B — covering sports stadiums, office buildings, and major infrastructure. Contract language evolved constantly through negotiations and legal amendments, making it nearly impossible for humans to assess risk exposure consistently. Ad-hoc spot analysis took weeks of manual effort, leaving underwriting, claims, and balance sheet positioning exposed to significant uncertainty.
What They Built
Mike's team at BCG mapped the contract value chain, identifying where human data entry introduced errors and where risk queries required hours of manual document review. They built a custom GenAI prototype on a RAG-based architecture, with semantic chunking, tagging, and hybrid SQL/document store retrieval to handle both structured data queries and open-ended risk questions. The system was trained extensively on reinsurance-specific legal language, with iterative QA using 10, then 100 contracts before scaling to the full 10,000. A custom LLM dictionary of reinsurance terms ensured domain-specific accuracy.
Mike's team began by mapping the contract value chain to identify where human data entry introduced errors and where risk queries consumed the most analyst time. The architecture challenge was significant: reinsurance contracts contained both precisely structured data — dollar amounts, dates, policy regions — and highly subjective legal language around inclusions, exclusions, and cyber clauses that resisted standard pattern matching.
To handle both query types, the team designed a hybrid retrieval architecture combining SQL-style structured queries for precise data extraction with semantic document store retrieval for open-ended risk questions. Semantic chunking was tuned to preserve legal clause boundaries. A custom LLM dictionary of reinsurance-specific legal terminology was built to ensure the model interpreted domain language consistently and accurately. QA was conducted iteratively — 10 contracts first, then 100, then the full 10,000 — allowing the team to identify edge cases and tune accuracy before scaling. The system reached 97%+ accuracy at production scale, using make.com and relevance.ai alongside the core RAG layer.
AI Role
The AI operates as a RAG-based retrieval and generation system trained on a custom dictionary of reinsurance-specific legal terminology, processing 10,000 complex contracts through semantic chunking, tagging, and hybrid SQL/document store retrieval. It answers both structured queries — such as counting contracts with specific clauses — and open-ended risk questions, generating responses grounded in the actual contract language with 97%+ accuracy.
AI Model
Custom / proprietary
Infrastructure
RAG architecture (hybrid SQL + document store) • make.com (workflow automation) • relevance.ai (AI orchestration layer) • Custom reinsurance legal LLM dictionary
Integration Points
Vector store connected to semantic chunking pipeline for 10,000+ contract documents • SQL layer integrated with structured contract data fields (amounts, dates, regions) • make.com automating query routing and output formatting • relevance.ai orchestrating LLM calls with domain-specific dictionary context
Impact
Speed — From Weeks to Hours
What previously required 2–3 weeks of manual analyst effort to complete ad-hoc contract risk spot analysis can now be done in less than a day by a single person, representing a dramatic acceleration in risk response time.
Accuracy — 97%+ Threshold Achieved
The system achieved and maintained 97%+ accuracy across 10,000 complex, 100+ page reinsurance contracts containing both highly specific structured data (dollar amounts, dates, regions) and ambiguous legal language around inclusions, exclusions, and cyber clauses.
Scalability and Reuse — Architecture Extends Beyond Contracts
The modular RAG architecture built for contract risk analysis is flexible enough to extend to claims data analysis and new contract negotiation support with minimal adjustments, providing compounding value from the initial investment.
Implementation Complexity
The solution required building a custom reinsurance legal dictionary, designing a hybrid retrieval architecture that handles both structured queries and open-ended risk questions, and achieving 97%+ accuracy across 10,000 complex 100+ page contracts containing highly ambiguous legal language. Iterative QA at scale — testing across 10, then 100, then 10,000 contracts — and the domain specificity of reinsurance legal language significantly increase implementation complexity.
Best Fit For
Mid-market to F100 companies in financial services, insurance (wealth management, underwriting, claims), and human capital management whose C-suite wants to connect AI investment directly to business strategy outcomes. Best suited for leaders with defined strategic priorities — growth or efficiency — who need a structured approach to identifying high-value AI use cases and managing the people/process changes required to realize ROI.
