Operations
SLA / SLO / SLI
Nested reliability commitments: an SLI (indicator) measures what actually happened, an SLO (objective) is the internal target, and an SLA (agreement) is the external contractual promise usually set below the SLO.
By Priya Ranganathan · Last updated July 22, 2026
In plain English
SLI = the number. SLO = the target you hit internally (e.g. 99.9% uptime). SLA = the promise in the contract (e.g. 99.5%). The gap between SLO and SLA is your error budget.
Example
Cloudflare's public SLA is 100% for uptime with credits kicking in on <99.9%. Their internal SLO is >99.99%. The SLI is measured minute-by-minute at every PoP.
Why it matters
Under-promising (loose SLA) with tight internal SLOs is the pattern that lets an engineering org sleep well AND keep enterprise customers happy. Reversed — public SLA tighter than internal capability — is how startups end up burning eng weekends refunding credits.
Common mistakes
- Setting SLA = SLO = SLI — no error budget for planned maintenance or gradual degradation
- Measuring only at the CDN edge — internal service SLOs are what actually degrade customer experience
- Not distinguishing availability from correctness (returning 200s with wrong data breaks both, only one shows in SLI)