Scaling Complex Financial Workflows with Agents

Legora, an agentic operating system for legal and professional services, has successfully integrated GPT-6 Astra to automate the labor-intensive process of financial-statement tie-outs. Traditionally, this task requires legal professionals to manually verify figures across draft accounts, trial balances, and consolidation schedules—a process that can consume entire evenings or even days. By deploying an agentic workflow, Legora reduced this to a single run that processes 41 documents in minutes, providing a granular, line-by-line record of every check performed.

Performance Gains and Human-in-the-Loop Design

Legora evaluated the model using their proprietary Benchmark for Agentic Reasoning (BAR), which tests performance on real-world legal tasks. The integration of GPT-6 Astra resulted in a 40% improvement in the financial-statement workflow compared to previous models. Key performance indicators included:

• Accuracy: The agent successfully identified 100% of the four planted errors, including a £500,000 discrepancy hidden within a revenue note.
• Completeness: The agent performed approximately 50 more checks than the previous model while maintaining 100% of the previous model's successful validations.
• Reliability: The system provides a comprehensive audit trail, allowing human experts to focus on final judgment calls rather than manual data verification.

This approach reinforces a "human-in-the-loop" architecture, where the AI handles exhaustive, repetitive data comparison, while legal professionals retain authority over the final decision-making process. This strategy is currently being expanded by Legora into broader professional domains, including audit, tax, compliance, and risk management.