Snowflake cuts contract review time 70% with AI agent

Snowflake's Forward Deployed Engineer team built an internal contract review agent to solve a scaling problem endemic to enterprise audit: as the company's deal volume grew — ranging from structured capacity commitments to specialized marketplace agreements — the audit team's traditional process of manually scanning PDF order forms line by line became unsustainable. Each contract carried unique commercial terms — capacity commitments, discount structures, billing frequencies, incentive clauses — that required careful review for revenue recognition compliance and internal audit control. Auditors sampled the population and tested controls across PDFs, but this approach did not scale to thousands of contracts per quarter.

Amrita Kapoor, VP Internal Audit, described the problem: "The audit team was spending hours manually scanning PDFs to find nonstandard terms and yet only providing assurance on a sample of the entire population. We needed to flip the model. By letting AI handle the exhaustive reading across the entire population, our auditors can focus their expertise on exception handling."

The Contract Review Agent operates in three layers. First, ingest and extract: Order form PDFs flow from source systems through Google Drive via Snowflake Openflow into a Snowflake stage. A Cortex Agent powered by Cortex AI Functions handles layout extraction and uses AI Extract to pull structured fields — customer name, capacity amount, discount terms, payment schedules, and dozens of revenue-relevant data points. Second, classify against a playbook and flag novel terms: Rather than relying solely on the model's judgment, the agent evaluates every extracted term against a user-managed playbook, a living set of rules stored in a Snowflake table that the audit team owns and edits directly with no engineering ticket required. The agent scores every clause, producing a classification of standard or nonstandard with confidence scores, clause-level excerpts, page references, and a natural-language explanation. The system also detects potentially novel terms by identifying clauses semantically distant from a corpus of known-standard contract language, then evaluates them against existing playbook rules to separate genuinely novel language from known patterns. Third, surface, review and learn: Findings are presented in a reviewer-centric application with an admin dashboard for portfolio-level visibility and a contract detail view for deep-dive review. Auditors can approve, override, or escalate every finding. Every correction — to an extracted value or a classification — is logged and persisted as extraction tips in long-term memory, fed into the agent on every subsequent run. A parallel Snowflake CoWork agent gives stakeholders a natural-language interface to audit status and nonstandard rates without needing direct app access.

Charles Xu, Engineering Manager of Applied AI, emphasized the reasoning layer: "The agent is not just parsing but reasoning. It explains why a term is considered nonstandard, cites the playbook rule and presents the contract excerpt, letting the reviewer confirm or correct. Every decision is logged."

The playbook manager puts control directly in auditors' hands. Each rule defines a term, its detection criteria, and its classification (standard vs. nonstandard). Every change is logged to an immutable audit trail showing action, rule ID, who changed it, when, old value, and new value. Rules take effect immediately on the next processing run without code changes or redeployment. This means the audit team can respond to a new contract pattern in minutes. When novel terms or discount structures appear, auditors add rules and the agent starts flagging them across the entire corpus. The novel term detection layer surfaces flagged terms in a separate view where auditors label them as meaningful anomaly or routine noise; these labels flow back into the system, teaching it what to prioritize in future runs.

The results speak to the impact of this design: review time was cut by 70%, what previously took multiple levels of prep across days is now handled by AI and takes hours. Thousands of order forms are reviewed for revenue implications per quarter without scaling the audit team in lockstep with deal volume. Because every extraction, classification, rule change, and correction is logged with full provenance, the system is auditable by design — the kind of evidence trail external auditors expect. And because each review cycle feeds corrections and labels back into the system, accuracy improves over time.

Nikolai Scholz, PM Unstructured Data Application, noted the broader implication: "The only way contract review scales with the business is through this kind of system. Without it, you grow your operational team linearly. With it, you keep the team lean while expanding coverage — and the system gets smarter every quarter because the experts are teaching it." The same architecture is being extended to other agreement types, generalizing the pattern: define an extraction schema, write playbook rules for what "standard" means in that document type, and let the agent do the exhaustive reading while experts focus on judgment.