Geometric means are often more useful than regular averages because they are closer to the middle value. When dealing with numbers that can only be positive, their distribution might not be normal but lognormal instead. If a normal distribution gives a standard deviation that is more than half of the average, it's better to use a lognormal distribution instead.