Back to glossary

SLI (Service Level Indicator)

A quantitative metric that measures the level of service being provided, serving as the concrete measurement against which SLOs are evaluated, such as request latency or error rate.

SLIs are the raw measurements that feed into SLO evaluation. Common SLIs include availability (percentage of successful requests), latency (request duration at a given percentile), throughput (requests processed per second), and error rate (percentage of failed requests). The SLI is the number you measure; the SLO is the target you set for that number.

Good SLIs are directly tied to user experience. Measuring server CPU utilization is not a useful SLI because users do not experience CPU utilization. Measuring the percentage of page loads that complete under 2 seconds is a useful SLI because it directly reflects what users experience. The closer the SLI is to the user's actual experience, the more useful it is for decision-making.

For AI features, effective SLIs might include time-to-first-token for streaming responses, percentage of responses that pass quality filters, embedding generation latency, and vector search recall accuracy. These measurements create a quantitative foundation for evaluating whether your AI features meet the reliability and quality standards your users expect.

Related Terms