I think something like backward induction could be used here to find a lower-regret but probably more complicated strategy: https://en.wikipedia.org/wiki/Backward_induction