87%
Request volume addressable by standard tiers
AI Governance Layer
Match AI capability to business risk, service need, and cost discipline.
The AI tier catalog is now route-native in `apps/web`, making governed AI design part of the real product shell instead of only a demo artifact.
3
Tiers allowed for routine request handling
100%
Risk-bearing actions requiring human approval
6%
Work expected to justify premium reasoning tiers
AI Tier Definitions
Service-oriented tiers, not vendor-branded model labels.
| Tier | Cost / Speed Profile | Best Fit | Guardrail |
|---|---|---|---|
| Tier 1: Lightweight Assist | Lowest cost, highest throughput, shallow reasoning | Knowledge lookups, form guidance, standard FAQ answers, request-field suggestions | No policy interpretation, no priority changes, no unsourced answers |
| Tier 2: Standard Service Copilot | Low-to-moderate cost, fast response, moderate reasoning depth | Request clarification, duplicate detection, triage summaries, draft status notes | May recommend but not decide routing, severity, or approvals |
| Tier 3: Governed Analyst | Moderate cost, slightly slower, stronger synthesis and comparison | Runbook step preparation, evidence comparison, problem clustering, change risk narratives | Requires approved sources, write-back logging, and human review before external impact |
| Tier 4: Specialist Operations Advisor | High cost, slower, deeper multi-step reasoning and expert-level analysis | Major incident support, vendor escalation packets, dependency-aware impact assessment, executive brief drafting | Use only for higher-severity work, material service risk, or leadership-level synthesis |
| Tier 5: Restricted Premium Specialist | Highest cost, lowest routine volume, deepest domain and synthesis capability | Cross-portfolio scenario analysis, complex root-cause synthesis, high-stakes policy support, sensitive multi-source reasoning | Formal approval required, restricted datasets only, no autonomous action of any kind |
How The Platform Chooses A Tier
Routing should reflect service type, urgency, risk, sensitivity, and cost discipline.
Incident impact
Higher-severity incidents move upward from routine assistance to governed operational reasoning.
Work class
Knowledge, intake, incident response, approvals, and leadership reporting should not use the same AI tier by default.
Decision exposure
If the output can affect service, access, policy, or public information, the routing model must shift to stricter guardrails.
Sensitivity
Restricted sources and regulated data narrow both the eligible tier and the approved source set.
Cost efficiency
When work is routine and low-risk, the system should route down to the least expensive safe tier.