Rollout Percentage

What it means

A rollout percentage is the fraction of eligible traffic that a flag turns on for. Set it to 5 percent and roughly one in twenty requests sees the new behavior while the rest fall back to the old path. The flag platform hashes a stable identifier, usually a user or account ID, so the same person keeps getting the same variant as you turn the dial up. Raising the number from 5 to 25 to 50 to 100 is what people mean by a percentage rollout or a canary release.

Why it matters

Shipping to everyone at once means any bug hits your whole user base before you can react. A percentage rollout caps the blast radius. If error rates or latency climb at 5 percent, you flip the flag off and only a sliver of users ever noticed. Because the assignment is sticky, users do not flicker between old and new on every request, which keeps the experience coherent and your metrics clean.

In practice

Most flag tools expose this as a slider or a targeting rule. ConfigCat offers percentage-based targeting on stable keys, and Harness FME builds staged percentage rollouts into its release workflow. The important detail is which attribute the hash is keyed on. Keying on user ID gives per-user stickiness; keying on a request with no stable ID randomizes every call and breaks the experience. A rollout percentage is also the foundation an A/B test sits on - once you can split traffic deterministically, you can measure the two arms against each other. See feature flag rollout strategies for how teams sequence the steps.