Benefit Street Partners uses Codex to cut credit analysis time from weeks to days

·2026.10.16 09:00

Key point

A portfolio manager built a BDC tracker in one day and analyzes 300-page credit agreements in 20 minutes using Codex.

Details

Benefit Street Partners (BSP), a credit-focused investment manager, is using Codex to accelerate underwriting and reduce the time required to answer investment committee questions from days or weeks to a single day. Portfolio manager Anant Kumar leverages the tool to build custom dashboards and analyze complex credit agreements, allowing the team to add review capacity and rigor without extending deal timelines.

Building Custom Tools with Codex

Kumar previously relied on software experts to create data dashboards. Using Codex for code generation and review, he built a tracker that compares loan valuations across business development companies (BDCs) using public SEC filings. The first version was completed in one day, with total development time, including refinements, estimated at 10–12 hours.

Automating Credit Agreement Analysis

The team also deployed a tool to analyze credit agreements, which often exceed 300 pages. By integrating memos from leading law firms on risk clauses, the analyzer produces a first-pass assessment in approximately 20 minutes. This is followed by 1+ hour of human review to ensure accuracy. BSP has refactored this tool with GPT-6 Astra and deployed it in a company-hosted portal.

Impact on Underwriting Workflow

The shift allows analysts to respond to committee follow-ups during meetings rather than waiting days for revised memos. Kumar notes that while AI assists with data collation and synthesis, final investment decisions remain with humans. The goal is not to increase deal volume, but to enhance the quality and speed of analysis within existing timeframes.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.