One sided test are better at detecting if parameter deviates in one or another direction. But with one sided test we loose ability to detect if parameter deviates in opposite direction.

Best approach is to always use two-sided test unless we clearly only care about parameter being differed in one direction (left or right) However, two-sided tests don’t tell anything about direction of sample. It only tell as whether sample is significantly different or not. It’s a bad practice to just say that sample is significantly different, conclusion also needs to include direction. (Confidence intervals)