New hybrid retrieval method balances accuracy and reliability in financial document AI systems by combining document routing with chunk-based search.

Traditional chunk-based RAG systems struggle with cross-document confusion in regulatory filings, leading to 22.5% failure rates versus 10.3% for document routing approaches

Semantic File Routing (SFR) improves robustness but reduces precision, achieving 6.45 average scores but only 8.5% perfect answers on the FinDER benchmark

Chunk-based retrieval (CBR) offers higher precision with 13.8% perfect answers but suffers from more catastrophic failures across 1,500 test queries

Researchers propose a Hybrid Document-Routed Retrieval method designed to resolve the robustness-precision trade-off in financial question-answering systems