I agree. I did some research out of curiosity in the past about multi-armed bandits. The optimal solution is quite involved, and not worth the effort of studying in my opinion.

Thompson sampling, on the other hand... I'm in love. Clean, simple, straight-forward, and so close to the optimal solution that you will never notice the difference. If you don't understand how to implement it, the stuff that you need to learn (mostly about the beta distribution) is generally useful and well worth studying. Though, after several years now, I'm still looking for an excuse to apply it somewhere. I can't get this method out of my head.