feature flags and experiments

Feature flags and experiments are related, but they serve different purposes.

Feature flags

A feature flag is a runtime switch that lets you turn a feature on or off without deploying new code.

Common uses

• Gradual rollout: release to 1%, then 10%, then 100%
• Kill switch: disable a broken feature instantly
• Environment control: enable in staging, disable in production
• User targeting: turn on for specific users, teams, or regions
• Operational control: separate deploy from release

Example

• new_checkout = true for internal users
• new_checkout = false for everyone else

---

Experiments

An experiment is a controlled test used to measure the impact of a change, usually by splitting users into variants.

Common uses

• A/B testing
• Product decisions
• Measuring conversion, retention, engagement
• Comparing multiple options

Example

• Group A sees old checkout
• Group B sees new checkout
• Measure whether new checkout improves purchase completion

---

Key difference

• Feature flag = control mechanism
• Experiment = measurement mechanism

A feature flag can be used to run an experiment, but not every flag is an experiment.

---

In practice

Many teams use the same infrastructure for both:

• A flag system decides who sees what
• An experiment framework logs exposure and outcomes

But they have different requirements:

• Feature flags care about safety, targeting, and rollout
• Experiments care about randomization, consistency, and statistical validity

---

Example architecture

• Flag service evaluates: “Should user 123 see feature X?”
• Experiment service assigns: “User 123 is in control or treatment”
• Analytics records exposure and results

---

Rule of thumb

• Use a feature flag when you want control
• Use an experiment when you want evidence

If you want, I can also give:

1. a comparison table,
2. examples in code,
3. or guidance on how to design a flag/experiment platform.