loading…
The dashboard says 99.9% HTTP success. Support says customers are receiving irrelevant answers and failed escalations.
HTTP 200 ≠ task succeeded
An SLI is a measured signal. An SLO is the reliability target for that signal. An SLA is the external commitment and consequence. Choose signals at multiple layers:
| Layer | Example signal |
|---|---|
| Infrastructure | availability, p95 latency, saturation |
| Integration | SDK request success by version, auth failures, contract errors |
| Workflow | payment-to-entitlement completion, duplicate side effects |
| Product outcome | correct resolution, successful escalation, task completion |
| AI quality | tool-selection accuracy, groundedness, policy compliance |
Choose three SLIs and give each an SLO. Include one infrastructure measure, one customer outcome, and one safety measure. Then state what engineering work should stop when the error budget is exhausted.
An availability SLO without an outcome SLO can hide a useless product. An outcome SLO without latency, safety, or cost can hide an unusable one.
Optional reading: Google SRE Workbook — Implementing SLOs.