The average? The median? The max? The min? The percentile? The standard deviation?
The problem is that there’s always a charitable message to be found when you cut the dataset in a certain way.
The hard part is having the discipline to stick to an objective measure even when the output looks dire.