An SLI is the measured user outcome, an SLO is its target over a window, and the error budget is the allowed bad fraction. Burn rate measures how quickly that budget is being consumed.
This guide defines each term, builds one request-based example, and turns it into multi-window alerts and a release policy using Google SRE's SLO alerting method.
What Are SLI, SLO, and SLA?
These terms are related but not interchangeable:
- SLI (Service Level Indicator): The metric you measure (for example, successful HTTP requests or p95 latency).
- SLO (Service Level Objective): The target for that metric over a time window (for example, 99.9% success over 30 days).
- SLA (Service Level Agreement): A contractual commitment, usually with financial consequences if missed.
Think of it this way:
- SLI = what you measure
- SLO = what you aim for
- SLA = what you promise externally
Most high-performing teams define internal SLOs that are stricter than public SLAs.
Why SLOs Matter for Monitoring
Traditional monitoring often answers: "Is it down right now?"
SLO monitoring answers a deeper question: "Are we delivering the level of reliability we promised over time?"
That difference is critical because many reliability issues are not full outages. They are slow responses, partial failures, regional issues, or elevated error rates that degrade user experience before a hard downtime event.
With SLOs, you can:
- Prioritize reliability work based on user impact
- Make safer release decisions
- Balance feature velocity against reliability risk
- Create clear escalation triggers
- Give leadership and customers objective reliability reporting
Picking the Right SLI
A good SLI reflects real user experience and is simple to measure consistently.
Common SLI categories:
Availability SLI
- Definition: Successful requests / total requests
- Example target: 99.9% over 30 days
Use this for APIs, checkout flows, auth endpoints, and any critical journey.
Latency SLI
- Definition: Percentage of requests under a threshold (or percentile target)
- Example target: 95% of requests under 500ms
Use this where "fast enough" matters as much as "up."
Quality SLI
- Definition: Correct responses / total responses
- Example target: 99.95% of responses pass content validation
Useful when 200 OK is not enough (for example, blank pages or incomplete API payloads).
Freshness/Timeliness SLI
- Definition: Time from event to processed state
- Example target: 99% of jobs complete within 2 minutes
Best for queues, async workers, and data pipelines.
How to Set a Practical SLO
Many teams either set SLOs too loose ("99% is fine") or unrealistically strict ("five nines for everything"). Use this approach:
- Start with user-critical paths
Define SLOs for login, payments, API read/write, and onboarding first. - Use historical data
Base targets on real baseline performance from your monitoring data. - Match business impact
Critical revenue paths deserve stricter targets than low-risk internal tools. - Choose a clear time window
7-day and 30-day windows are most common for operational decisions. - Keep it understandable
If non-engineers cannot interpret the objective, refine it.
Error Budgets: The Decision Engine
An error budget is the amount of unreliability you can "spend" while still meeting your SLO.
If your availability SLO is 99.9% over 30 days, your monthly error budget is 0.1%.
That budget creates explicit trade-offs:
- Budget healthy: Continue shipping features at normal pace
- Budget burning fast: Slow high-risk changes and increase safeguards
- Budget exhausted: Pause risky releases, focus on reliability work
This replaces subjective arguments with measurable policy.
SLO Burn Rate: Catch Problems Early
SLO breach alerts that trigger only at the end of the month are too late. Burn rate alerts detect when you're consuming error budget too quickly.
Example:
- 30-day SLO: 99.9% availability
- You can tolerate 0.1% failure over 30 days
- A 2% bad-event rate consumes a 0.1% error budget at 20� burn rate
Use paired windows so an alert requires both meaningful budget consumption and evidence that the problem is still active. Google SRE's worked example pages at 14.4� over both 1 hour and 5 minutes; the shorter window is roughly 1/12 of the longer one. Add a slower ticket path for chronic burn. Treat these as a starting pattern and calculate thresholds from your own SLO window and budget policy.
SLI/SLO Examples by Service Type
| Service Type | Suggested SLI | Example SLO |
|---|---|---|
| Public website | Successful HTTP checks | 99.9% over 30 days |
| Auth API | Successful login requests | 99.95% over 30 days |
| Checkout | Successful completed transactions | 99.95% over 30 days |
| Internal dashboard | Successful page loads | 99.5% over 30 days |
| Queue workers | Jobs completed within threshold | 99% within 2 minutes |
| Webhook delivery | Successful callback delivery | 99.9% over 7 days |
Start with a few SLOs you can operationalize. You can expand coverage over time.
Common SLO Mistakes
Measuring system health, not user experience
CPU and memory are useful diagnostics, but poor primary SLIs. Your SLO should map to what users care about.
One SLO for everything
Not every endpoint has the same criticality. Use tiers.
No response policy
An SLO without an action policy is just a dashboard number. Define which burn conditions page, create a ticket, pause risky releases, require a postmortem, or trigger reliability work. The Google SRE example error-budget policy shows how to connect budget exhaustion to release decisions.
Ignoring planned changes
Major migrations and launches can consume budget quickly. Plan with temporary safeguards and tighter monitoring.
Overcomplicated formulas
If engineers debate math more than reliability outcomes, simplify.
Implementing SLO Monitoring with Webalert
Webalert gives you the external signal needed for practical SLO monitoring:
- Frequent checks (1-minute intervals) for accurate availability measurement
- Multi-region monitoring to reflect real user reachability
- Response time tracking for latency-oriented objectives
- Content validation to ensure responses are correct, not just reachable
- Heartbeat monitoring for background jobs and pipeline freshness goals
- Incident timelines to quantify budget consumption and trend reliability
- Alert routing via Email, SMS, Slack, Discord, Teams, and webhooks
With these signals, you can calculate objective reliability targets and enforce error-budget policies with confidence.
See features and pricing for details.
Getting Started This Week
If you are new to SLOs, keep scope tight:
- Pick one critical user flow
- Define one availability SLI
- Set one 30-day SLO
- Add burn-rate style alerts
- Review weekly with product + engineering
Within a month, you will have clearer reliability decisions and fewer debates driven by gut feeling.
Summary
- SLI is the measured reliability metric.
- SLO is the target for that metric over time.
- Error budgets convert reliability targets into release decisions.
- Burn-rate alerting catches risk before you fully breach.
- User-centric signals (availability, latency, correctness) are the right SLO foundation.
- External monitoring is essential for objective SLO tracking.
SLO monitoring does not replace incident response. It improves it by making reliability measurable, actionable, and aligned with business priorities.