A Service Level Indicator, or SLI, is a precise measurement of how a service is actually performing. It is a number you can track over time, such as the percentage of successful requests, the average response time, or the share of pages that load within a target window. The SLI is the raw signal of reliability.
Think of it like the speedometer in a car. It does not decide how fast you should go; it simply reports how fast you are going right now. An SLI works the same way: it reports the real-world behaviour of your system, drawn from the metrics your monitoring collects. Good observability is what makes accurate SLIs possible.
On its own an SLI is just a fact. It becomes useful when paired with a target. That target is the SLO, and when the SLO is written into a customer contract it becomes an SLA. Choosing the right SLI is harder than it sounds, because the obvious number is not always the honest one. Take average response time. It can look healthy while a slow tail of requests quietly ruins the experience for one user in twenty. So teams often track a percentile instead, like the slowest 5 percent of requests, which is closer to what a frustrated person actually notices. The best SLIs sit close to the user. The share of checkouts that complete, the share of searches that return in under a second. These tell you whether the product is working, not just whether a server is technically up.
At TopDevs we pick a handful of SLIs that reflect what users actually feel, so a client measures real experience rather than vanity numbers that look healthy but mean little.
