Tools

AI-generated text

Legora’s Agent with GPT‑6 Astra reconciled 41 financial documents in minutes, finding planted errors

Legora used GPT‑6 Astra in its agentic platform to perform a financial‑statement tie‑out across 41 documents in a single run, completing the task in minutes and surfacing planted errors including a £500,000 gap.

Legora’s Agent with GPT‑6 Astra reconciled 41 financial documents in minutes, finding planted errors

Legora’s agentic operating system, using the GPT‑6 Astra model, completed a financial‑statement tie‑out across 41 documents in a single agent run, which the company says took minutes. During the process the system detected intentionally planted errors, including a £500,000 gap hidden in a revenue note.

What the workflow involved and why it matters

A financial‑statement tie‑out requires checking every figure in draft accounts against trial balances, a consolidation schedule, and the previous year’s accounts until items reconcile. Legora Legal Engineer Percevale Perks notes that this work "can take an entire evening, sometimes days." Because the task is repetitive and time‑consuming, automating the exhaustive comparisons can speed up the first pass and reduce manual effort while leaving final judgments to experts.

Performance and findings

  • The Legora Agent, powered by GPT‑6 Astra, reviewed 41 documents in one run, checking each balance against its supporting schedule and recording each check.
  • Legora reports the Agent completed the task in minutes and surfaced breaks in the amounts for review.
  • On Legora’s internal Legora Benchmark for Agentic Reasoning (BAR), GPT‑6 Astra delivered almost a 40% improvement on this specific financial‑statement workflow compared with the previous model; across all BAR tasks the average improvement was about 3%.
  • In the tie‑out test the model found all four errors Legora had planted in the accounts, including a £500,000 discrepancy in the revenue note.
  • The Agent retained every check the previous model had completed correctly and performed around 50 additional checks.

Human oversight preserved

Legora emphasises that the Agent performs the exhaustive comparisons and produces a granular record of every line item and figure, but the legal or financial expert remains responsible for the final judgment on each result. This human‑in‑the‑loop approach underpins Legora’s plans to extend the platform beyond legal work into audit, tax, compliance, and risk workflows.

Scale and customer base

Legora says its platform is used by more than 100,000 professionals across over 1,800 in‑house legal departments and law firms in more than 50 markets. Its legal engineers work directly with customers to adapt Legora to end‑to‑end workflows, from contract review to legal research.

Conclusion

Integrating GPT‑6 Astra into the Legora Agent produced a faster, more complete first pass of a financial tie‑out and a clearer audit trail for review, while maintaining that final decisions remain with legal professionals.