When should we bother with a feature flag system versus just redeploying?

I'm trying to figure out when to use a whole feature flag system instead of just putting out a new version of our service. For little changes, it just feels like too much work to add a flag, deal with its state, and then clean it up later. But for bigger features, or stuff we want to A/B test, it seems super important. So, how do you decide when to actually use a feature flag? Are there certain changes where you always go for them, or times when you just skip them altogether?