​

SLO Playbook

​

1. Pick the Right SLI First

• Start with customer journey (sign-in, run workflow, call agent).
• Choose indicators: latency p95, error rate, freshness, throughput.
• Validate data availability before committing.

​

2. Craft the Objective

• Format: “Measure ≤ Target for % of requests over rolling window”.
• Example: “Productivity automations p95 latency <= 700 ms for 99% of runs in a 28‑day window”.
• Tie objective to user outcome (fast automations, reliable agents, etc.).

​

3. Error Budgets & Guardrails

• Error budget = 1 - SLO (e.g., 1%).
• Track burn rates (1h, 6h, 24h). Alert roles when exceeding policy.
• Link features/experiments to budget consumption.

​

4. Review Cadence

• Weekly SLO review: status, burn, actions.
• Monthly: re-evaluate if SLO still meaningful; adjust targets based on data + business goals.
• Post-incident: decide whether threshold/indicator needs revision.

​

5. Tooling

• Nobl9 / Sloth / custom calculators generate SLO configs for Prometheus.
• Grafana dashboards with error budget viz + annotation for deployments.
• Slack bot posts SLO digests with CTA to relevant runbooks.

​

6. Templates

### SLO Card
- Service:
- Indicator:
- Target:
- Window:
- Alert policy:
- Owners:
- Links: dashboards, runbooks, repos