Graham McNicoll

A lot of engineering teams have feature flags to control the release of features and act as a circuit breaker. If something goes wrong, they just turn it off. The frustrating part is that the natural next question should be: okay, if it passed the circuit breaker, what did it do to my metrics, what did it do to my users. And surprisingly few engineers actually ask that. The thing is, your feature flag is already your decision engine. It's already doing assignment to your users. So asking it to randomly assign people into a control and a variant is trivially easy. If you have feature flags, you're so close to running a world-class experimentation program. In GrowthBook it's literally a few clicks. If you haven't used feature flags and experimentation together, I'd encourage you to give it a try. It's really transformative to how you ship and release features.