Q: Your company's SLO for an API is 99.95% availability. The CTO asks, "Why not just set it to 99.99%?" What is the SRE argument against over-committing to higher SLOs?

Higher SLOs are exponentially more expensive to achieve, and over-committing creates perverse incentives.

• Redundant everything: Multi-region active-active, automated failover, zero-downtime deployments
• Zero human-in-the-loop: 4.3 minutes is too short for humans to detect, diagnose, and fix—everything must be automated
• Dependency chain ceiling: If your database (e.g., RDS) offers 99.95% SLA, your service physically cannot achieve 99.99% unless you build multi-database redundancy
• Development velocity plummets (every change is a risk to the SLO)

Increasing API availability from 99.9% to 99.99% reduces the allowable monthly downtime from ~43.8 minutes to just ~4.3 minutes—a 10x reduction in error budget. This leap requires fundamental architectural shifts from single-region resilience to multi-region active-active clusters with automated sub-minute failover.

• Exponential Cost Curve: Moving from three nines to four nines typically triples infrastructure costs due to multi-region data replication and networking egress.
• SRE Principle: Align SLOs with actual user pain thresholds; pursuing four nines when users are satisfied with three nines starves feature velocity.