Requests classifiable without manual intake rework
Choose the right AI before the request touches the model.
The routing policy now lives inside the production route shell, making AI governance part of the real frontend path instead of only a side demo page.
Demand kept on lower-cost eligible tiers
Requests requiring approval before AI output is used
Demand routed away from AI due to policy or risk
Routing Decision Matrix
The engine makes tier decisions by service pattern, not by user guesswork.
| Work Pattern | Default Route | Escalation Trigger | Human Requirement |
|---|---|---|---|
| Knowledge and FAQ guidance | Tier 1 lightweight assist | Low confidence, missing source, or policy interpretation request | Escalate to service desk or knowledge owner when unsupported |
| Request intake clarification | Tier 2 standard service copilot | Approval-bound request, restricted data, or unclear business owner | Requester confirms summary before submission |
| Approval-bound access request | Tier 2 for field readiness only | Privilege level, exception pattern, or missing approver | Manager or security approver must authorize next step |
| Incident evidence preparation | Tier 3 governed analyst | Severity 1-2, cross-system blast radius, or vendor escalation | Incident lead validates evidence and decisions |
| Major incident command support | Tier 4 specialist operations advisor | Public communication, failover recommendation, or sensitive data scope | Commander retains all send and action authority |
| Leadership scenario analysis | Tier 4 or Tier 5 by approval | Cross-portfolio consequences, sensitive policy interpretation, or premium-model request | Formal sponsor approval for Tier 5 use |
Policy Flow
Every request should move through the same four routing decisions before AI output is trusted.
Classify
Identify service family, category, urgency, requester role, and approval dependency.
Check Risk
Apply sensitivity, public impact, privilege, and service-risk checks.
Select Tier
Choose the lowest-cost AI tier that still meets accuracy and governance needs.
Gate Output
Require confirmation, escalation, or full human takeover when output can change service or policy posture.