The average is the least interesting number
What a mean hides, and which statistic to ask for instead.
The average is the least interesting number in a data set because it collapses everything about how spread out, skewed or clustered the underlying values actually are into a single figure, discarding exactly the information most likely to matter for a real decision.
Huff's How to Lie with Statistics, from 1954 and barely a hundred pages long, is probably the fastest way there is to become properly suspicious of a graph.
One hand in ice, one in boiling water
A person with one hand in a bucket of ice water and the other in a bucket of nearly boiling water is not comfortable, even though the average temperature across both hands lands on a pleasant, room-temperature figure. Neither hand is anywhere near that number, and the comfortable-sounding average exists nowhere in the experience being measured. The example is old and deliberately absurd, which is what makes the flaw impossible to miss, and the same flaw sits quietly inside many ordinary-looking reports where nothing about the number signals that anything has been hidden.
Three data sets with the same mean
An average is calculated by adding every value together and dividing by how many there are, and that calculation treats every value the same whether it sits close to the pack or a long way out on its own. One data set might have every value clustered tightly around its mean. A second might have half its values far below and half far above, with barely any near the middle. A third might have almost every value near the middle and one or two far away, dragging the calculation just enough to land on the same figure. The arithmetic has already thrown away the shape of the data by the time it produces its answer, so a reader handed only the average cannot tell which picture produced it.
That matters most where a decision hinges on individual cases near an extreme. A batch of parts whose average dimension sits dead on the design target might contain a large share running close to a tolerance limit, or almost none. A production line whose average cycle time is on target might be steady, or lurching between very fast and very slow runs that merely balance out.
Asking for the spread, or the median
Asking for the spread alongside any average, as a range, a standard deviation, or simply a look at the actual distribution, turns a number that hides its own reliability into one that reveals it. Where a few extreme values pull the average away from where most values sit, the median (the middle value once everything is sorted) is often the better summary, since one outlier barely moves it. In the set 1, 2, 3, 4 and 90, the mean is 20 while the median is 3, and only the median describes a typical member.
The average still earns its place when the data are tightly clustered, where a spread figure mostly confirms what the mean implied. It is also the right figure whenever a total matters more than any individual case, because extremes wash out in a sum. The caution belongs to decisions about parts near a tolerance limit, components near a thermal limit or patients near a dangerous reading, where the average's blindness to shape hides the very values the decision needed to see.