Document Processing · Risk Management
Sullivan & Cromwell needed to review draft agreements against relevant precedents, client priorities, and market practice in one workflow so lawyers could do deeper legal analysis without losing transparency into sources, uncertainty, or final human judgment.
Reported impact
Agreement Analyzer helps Sullivan & Cromwell lawyers identify potential issues, compare terms across precedents, produce proposed changes with visible supporting authority, and spend more time on the legal questions where human judgment matters most while delivering a higher-quality product for clients.
In collaboration with OpenAI, Sullivan & Cromwell built Agreement Analyzer to concentrate agreement review in a single AI-assisted workflow. A lawyer inputs a draft agreement, then adds selected precedents, a negotiating playbook with firm and client positions, and market-practice context. Agreement Analyzer uses ChatGPT Enterprise to run multiple review cycles from different perspectives, compare corresponding provisions across precedents, surface clause interactions, and synthesize the findings into a memorandum plus proposed tracked changes. The implementation emphasizes transparency rather than autonomous action: supporting authority and sources remain visible so claims about market practice can be inspected, refined, accepted, or rejected by a lawyer. The system evolved from a single-user proof of concept into a parallelized, multi-lawyer tool that brings more of the firm's internal expertise into complex agreement analysis while keeping lawyers responsible for final judgment.
A lawyer uploads a draft agreement into Agreement Analyzer, adds selected precedents plus a negotiating playbook with firm and client positions and market-practice context, and sends the package through multiple ChatGPT Enterprise review cycles. Agreement Analyzer synthesizes the results and returns a memorandum and proposed tracked changes with supporting authority for a lawyer to inspect, adapt, or reject.
"Agreement Analyzer integrates Sullivan & Cromwell's workflows, curated firm precedents, negotiating playbooks and ChatGPT Enterprise."
"If an AI tool makes a claim about market practice, you need to be able to see the source of that information."
"Sullivan & Cromwell partner John Savva developed Agreement Analyzer as a single-user proof of concept built with Codex."
"The firm's Agreement Analyzer brings those inputs into a single review workflow, producing analysis and proposed changes that lawyers can inspect and refine."
"The tool performs multiple review cycles from different perspectives, synthesizes the outputs, helps identify potential issues and returns a memorandum and proposed tracked changes for a lawyer to review, adapt or reject."
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
OpenAI Sample Oct 16, 2026 Benefit Street Partners Benefit Street Partners needed to answer investment-committee follow-up questions faster, compare lender marks across shared loans, and review 300-page credit agreements for clauses that could shift value away from… 10–12 hours total. BDC mark tracker · Codex · Credit agreement analyzer 
Anthropic Sample Archived · Sep 15, 2026 DXC Technology DXC needed to reduce insurance document backlogs and automate highly regulated underwriting and claims decisions without losing auditability, policy fidelity, or human oversight. 8 hours to stand up a regulated benefits calculation, 80% right on the first pass Assure · Claude Code · Claude Platform 
Alibaba Cloud Sample Jan 6, 2026 Chengdu Wuniu Technology (成都午牛科技有限公司) Wuniu was handling messy blue-collar recruiting data with manual teams and brittle keyword rules. Resumes were informal and hard to parse, job postings often bundled multiple roles or contained compliance risks, and… extraction accuracy improved about 40% Alibaba Cloud Bailian · Qwen-Plus 