Just built a feedback loop where analysts vote 'useful/not useful' on every AI suggestion.

I've been lurking here for a while, learning a ton. I work on the process side of our SOC.

We just implemented a simple feedback loop in our AI-assisted triage system. For every alert summary or investigation step the AI suggests, the analyst has to click 'useful' or 'not useful' before moving on.

It's only been a week, but the data is already surprising. The 'not useful' votes are clustering around specific alert types. It feels like we're manually training it on the fly.

Has anyone else tried something like this? I'm curious about two things: how you structured the feedback (binary vs. more granular), and if you've connected it back to retrain or fine-tune your models yet.