Staggered Rollout

When a company has finished a new version of its app or program, it could ship it to all users at once. In a staggered rollout, it does exactly not do that. Instead, only a few users get the new version first, say one percent. If everything runs smoothly there, the share is increased: five percent, twenty percent, and eventually everyone. The name comes from the English word “to stagger,” meaning to arrange things at intervals. It’s also called a staged or phased rollout.

Why nobody delivers to all users at once anymore

Software always contains bugs that nobody found beforehand. In-house tests run on selected devices and with hypothetical users. Reality is messier: thousands of phone models, slow networks, odd settings. A bug often only shows up there.

This is exactly where the benefit of staggering lies. If a version only goes out to one percent and crashes there, 99 percent of users are unaffected. The company pulls the version, fixes the bug, and starts again. With a release to everyone, this would have become a headline.

For publicly traded tech companies, this is also a matter of money. An outage in an online shop’s checkout process costs measurable revenue per hour. Recalling a version that has already been distributed worldwide is also slow, because many devices don’t reload immediately. That’s why the staggered rollout is now standard at almost all major software companies.

From one percent to a hundred: the stages and their brake

Technically, this requires two things. First, a way to divide users into groups, usually at random based on a device or account identifier. Second, a switch within the program that turns a feature on or off without requiring a new version to be installed. Such switches are called feature flags.

While the rollout is running, the team monitors metrics. Typical ones are crash rate, load time, server error messages, and whether users close the app faster than before. The new group is always compared to the large remainder that still has the old version. This remainder acts like a control group in an experiment.

If a metric deteriorates significantly, the rollout is stopped or rolled back. Many companies automate this: if the crash rate exceeds a threshold, a program automatically switches the feature off again. This is called an automatic rollback. It’s important to distinguish this from an A/B test: there, the goal is to find out which of two variants is preferred. With a staggered rollout, the decision has already been made; the only thing being checked is whether the release is safe.

How to tell you’re caught in a rollout stage

You’ve probably experienced this already. A friend shows you a new feature in Instagram or WhatsApp, but it’s missing on your phone, even though both phones display the same version. In that case, the feature is switched on for them via a flag and still off for you. That’s exactly why checking the App Store for an update doesn’t help in this case.

The app stores themselves also offer this. On Google Play, a developer can send an update to a freely chosen percentage of users. Major operating system updates from Apple or Microsoft also appear in waves, often first in individual countries or for certain device models.

In the news, the term frequently comes up with AI products. New chatbot models usually launch with paying customers or in a few countries first, before becoming available everywhere. Phrases like “will be rolled out over the coming weeks” describe exactly this. For investors, the stage is interesting because a rollout that stalls at ten percent often points to a technical or legal problem.