Was this analysis based on testing only two variations, A vs. B, using A/B testing then MAB?
If so then that ignores one of the best advantages of MAB is you can compare variations A, B, C, D, E, F, ... all simultaneously, where if using A/B testing you'd have to coordinate a tournament between all of those variations to determine the best of the bunch ... so seems it would take just as long with A/B testing (as thus a large sample population) to arrive at knowing the best variation from many choices.
If so then that ignores one of the best advantages of MAB is you can compare variations A, B, C, D, E, F, ... all simultaneously, where if using A/B testing you'd have to coordinate a tournament between all of those variations to determine the best of the bunch ... so seems it would take just as long with A/B testing (as thus a large sample population) to arrive at knowing the best variation from many choices.