Prioritizing Service Level Indicators Over Objectives for Effective Reliability

Why SLIs Matter More Than SLOs

Samson Tanimawo, CEO of Nova AI Ops, asserts that technical teams frequently prioritize arbitrary targets over accurate measurement. He argues that an SLO is merely a decision, while the SLI represents the actual signal of user experience.

Why This Matters

In technical environments, teams often focus on vanity metrics like 99.9% uptime for healthcheck endpoints, which creates a false sense of security. If the underlying SLI does not capture the user’s actual journey—such as checkout completion within 5 seconds—the resulting SLO targets become meaningless, leading to on-call fatigue without resolving real-world service degradation.

Key Insights

• SLOs are arbitrary numerical decisions, such as 300ms p95 latency, whereas SLIs are the fundamental signals being measured.
• Healthcheck endpoints returning 200 OK are poor SLIs because they do not guarantee the functionality of the actual API or product.
• Effective signals, such as user-initiated checkout success rates, provide high-fidelity data regardless of the specific target percentage chosen.
• The On-Call Test determines SLI quality: if a missed SLO doesn’t correspond to user suffering, the measurement signal is incorrect.

Practical Applications

• Use case: Monitoring checkout requests with Nova AI Ops to ensure successful completion within a 5-second threshold. Pitfall: Using shallow healthchecks that mask backend API failures.
• Use case: Defining reliability targets based on user-facing latency rather than internal system uptime. Pitfall: Gaming metrics through aggressive caching that hides real service latency.

References:

• https://dev.to/samson_tanimawo/why-slis-matter-more-than-slos-34jp
• https://novaaiops.com