Service levels for your AI systems
A service level is what we commit to once your system is live: how fast we respond by severity, when someone is reachable, and who you escalate to if a response is late. These figures are named in the contract, not left as an intention.
Response-time tiers
| Severity | Description | Response time |
|---|---|---|
| P1, critical | The system is down or producing answers a person could act on wrongly. Business-critical impact. | < 30 minutes |
| P2, degraded | The system is running but with reduced accuracy, latency or partial availability. | < 2 hours |
| P3, general | Questions, minor issues, or requests that do not affect production use. | < 4 hours |
Service hours
Standard coverage
Monday to Friday, during normal business hours in the relevant time zone. The default at the Basic and Extended tiers.
Wider coverage, by agreement
Extended hours or a defined on-call window for systems where downtime outside business hours still carries a cost.
If a response is late
Named contact
Every customer has a named contact for their service tier, not a shared inbox.
Deputy contact
At the higher tiers, a named deputy is reachable if the primary contact does not respond.
Escalation to management
Unresolved P1 issues escalate to a named manager, with a fixed time limit for that step.
Written follow-up
A short written note on cause and resolution, so the same issue does not repeat unnoticed.