Expense Policy Rule

Expense policy is whatever the approving manager thinks is reasonable. One manager approves $500 dinners, another rejects $15 parking charges. When a new employee asks 'what's the per-diem for Chicago?' nobody can point to a written answer. The policy is 'ask your manager' and the answer depends entirely on who you ask.

AI cannot review expense reports because no rules exist to review against. Every expense requires a human judgment call with no standard to apply.

An expense policy document exists — a PDF in the employee handbook from three years ago. It says 'economy class for flights under 4 hours' and 'meals up to a reasonable amount.' But 'reasonable' isn't defined, the per-diem rates are outdated, and the document doesn't cover remote work expenses, rideshare services, or co-working space rentals. Approvers interpret the policy differently.

AI could reference the policy document for basic rules, but cannot apply it consistently because key terms are ambiguous ('reasonable'), rates are outdated, and common expense categories aren't addressed.

Expense policy rules are documented in a comprehensive policy manual with specific thresholds — per-diem rates by city, meal limits by meal type, hotel maximums by tier, travel class by flight duration and employee level. Approval levels are defined by dollar amount ($0-500 manager, $500-2000 director, $2000+ VP). But the rules live in a PDF that the expense system doesn't read — humans apply them manually during review.

AI can flag expenses that obviously exceed documented thresholds (e.g., hotel rate vs. city maximum) but cannot enforce nuanced rules because the policy document isn't structured in a machine-readable format.

Expense policy rules are encoded in the expense management system with machine-readable logic. Per-diem rates are lookup tables by city and date. Approval routing is automated based on amount, category, and submitter level. Required documentation (receipt, justification, pre-approval) is enforced by category. An analyst can query 'what is the current hotel limit for a director traveling to San Francisco?' and get a definitive answer.

AI can automatically review most routine expenses against encoded policies, flag violations, route approvals, and enforce documentation requirements. Cannot handle edge cases where policy intent requires judgment (e.g., 'client entertainment at a sporting event — is this meals or entertainment?').

Expense policy rules are fully schema-driven with formal entity relationships. Each rule links to applicable employee populations, geographic rate tables, approval hierarchies, documentation requirements, and exception criteria. Edge case classification rules handle ambiguous categories. An AI agent can evaluate 'this employee is a VP traveling to Tokyo for client meetings — what are all applicable expense limits, documentation requirements, and approval paths?' with a structured, complete answer.

AI can autonomously review and approve routine expenses, classify edge cases using formal rules, recommend policy exceptions with supporting rationale, and generate compliance reports. Human review required only for truly novel expense categories.

Expense policy rules are dynamic and self-adjusting. Per-diem rates update automatically from market data sources (hotel rate indices, GSA tables, cost-of-living feeds). Approval thresholds adjust based on organizational changes. New expense categories (AI subscriptions, EV charging, home office equipment) are classified automatically using pattern recognition against the policy ontology. The policy adapts to reality, not the other way around.

Fully autonomous expense policy management. AI maintains, enforces, and evolves expense rules in real-time based on market conditions, organizational changes, and emerging expense patterns.

Ceiling of the CMC framework for this dimension.