Agentic Intelligence for Contract Review on Snowflake

Snowflake’s Forward Deployed Engineer team developed an internal AI-powered contract review agent using Snowflake Openflow, Cortex AI, and Streamlit to automate the review of customer contracts. The tool processes high volumes of order forms, MSAs, and amendments to identify nonstandard clauses that impact revenue recognition and audit controls. By automating the detection layer, the agent aims to eliminate the operational bottleneck caused by manual review while maintaining auditor oversight over what constitutes nonstandard terms.

The agent extracts structured data from PDFs, including customer names, capacity amounts, discount terms, and payment schedules, using Snowflake AI Extract and Cortex Agent. It then classifies each extracted term against a playbook—a governed Snowflake table managed by the audit team—to flag nonstandard clauses with confidence scores, excerpts, and explanations. Findings are surfaced in a reviewer-centric application, allowing auditors to approve, override, or escalate each flagged term while logging corrections to refine future extractions.

A parallel Snowflake CoWork agent provides stakeholders with a natural-language interface to audit status and nonstandard rates without direct app access. The system uses a two-layer approach to detect novel terms: semantic distance from a corpus of standard language and evaluation against playbook rules. Flagged novel terms are reviewed by auditors, who label them as meaningful anomalies or routine noise, feeding back into the system to improve future detection accuracy.

The architecture is designed to scale with business growth by automating exhaustive contract review while keeping auditors focused on judgment. Nikolai Scholz, PM Unstructured Data Application, notes that this system allows teams to expand coverage without linearly increasing headcount, with the agent becoming more precise as experts refine its playbook and rules over time.