# AI service SLO template

Use this template to operate AI workflows as services instead of demos.

## SLO dimensions

| Dimension | Example target |
| --- | --- |
| Availability | Service responds or degrades gracefully during provider issues |
| Latency | P95 response time by workflow and model route |
| Quality | Task success, citation quality, and reviewer override rate |
| Cost | Cost per successful task and cost drift threshold |
| Safety | Policy violation rate and approval bypass rate |
| Support | Escalation volume and time to resolution |

## Operating rules

- Define degraded mode before providers fail.
- Alert on workflow success, not only infrastructure health.
- Review SLO misses alongside eval failures and incident reports.

## Output

Observable targets, dashboards, alerts, and escalation rules for every production AI workflow.