Service level indicators

Give the service objective an owner who can change priorities

Reliability reporting matters when someone can act on it. Hand over the definition, response policy and authority to choose corrective work.

In this article

Explain the measure with examples

Provide the eligible population, good outcome, window and exclusions. Include a few concrete operations showing how success, failure and late completion are counted.

Link the implementation query and its data source. The receiving team should be able to explain a result without relying on the dashboard author's memory.

Record known coverage gaps, including failures the measurement cannot observe.

Assign the decision responsibility

Name the service owner who can balance reliability work with other delivery priorities. Separate that responsibility from maintaining the monitoring pipeline.

A team can keep charts running while no one acts on repeated missed objectives. The operating policy should explain what review or change follows a sustained problem.

Avoid promising automatic product decisions from a number alone. The owner still needs context about impact, cause and available improvements.

Hand over the incident path

Show how an alert leads to affected operations, diagnostic evidence and a tested recovery procedure. Include who is contacted when telemetry itself is missing.

Ask the receiving operator to investigate a known synthetic failure. Observe whether the documentation connects the percentage to a concrete user journey.

Fix missing access or unclear classifications before closing the handover.

Keep the definition current

Assign a review when the application changes its workflow or architecture. Moving work into a queue, adding a new client or changing eligibility can alter the measure's meaning.

Preserve definition versions and annotate transitions in reports. Historical comparisons should not conceal a changed denominator.

Schedule a regular service review that turns findings into owned work. The handover is complete when the team can interpret the evidence, respond to current impact and make informed reliability decisions, rather than merely acknowledge that a dashboard exists.

Trace one export into the indicator

Give the incoming team one event that should count as good, one that should count as bad and one excluded by the definition. For an export service, a returned job ID may start the operation while a verified downloadable file completes it. A cancelled request might be excluded under an explicit rule rather than counted opportunistically.

Have the receiver locate those events in the actual query. This checks that the written definition and dashboard still describe the same business outcome after instrumentation or schema changes.

Primary sources

Google SRE: implementing service objectives

References checked 11 September 2026.