You see it in performance reviews. You see it in hiring rubrics, salary bands, school curricula, diet guidelines, public health messaging. The methodology is always the same: aggregate the data, find the center, build the system around the center. Clean. Defensible. Catastrophically wrong.
Here is the problem. The center does not exist.
Francis Galton invented the statistical concept of regression to the mean in 1886 to describe how extraordinary parents produce more ordinary children. The discovery was real. The misapplication came fast. Within decades, the average had become a design standard, as if the mathematical artifact of central tendency told you something about the actual person sitting in front of you. It does not.
Todd Rose documented this in The End of Average with a case study the U.S. Air Force ran in the 1950s. They were losing pilots. Too many accidents. Their hypothesis: the cockpit was designed for the average 1926 pilot, and the 1950 pilot had a different body. So they measured 4,063 pilots across ten dimensions, chest circumference, thumb length, sitting height, and calculated the average for each. Then they checked how many pilots actually fit within 30% of average across all ten dimensions simultaneously.
Zero. Out of 4,063 pilots, not one was average.
The cockpit was designed for a person who did not exist. The system optimized for a fiction and then blamed the individual when the fiction failed them.
That is the pattern. In schools, the curriculum is pitched at the median student, too fast for some, too slow for others, genuinely right for very few. In medicine, drug dosages are calibrated on average body weight and average metabolic rate. In management, performance frameworks compress individuals into rating bands that obscure both the ceiling and the floor. The average becomes a ceiling disguised as a floor.
What gets lost is the shape of the distribution.
Averages flatten variance, and variance is where the interesting information lives. In a population of ten people with a mean score of 70, you cannot tell whether you have ten people who scored 70 or two people who scored 40 and 100 respectively. The strategies, the diagnoses, the interventions are entirely different. The average tells you nothing about which problem you are actually solving.
This is especially destructive at the tail. Genuinely exceptional performance, whether in athletics, research, art, leadership, does not look like an elevated average. It looks like a different shape altogether. Michael Jordan's shot selection, Feynman's pattern recognition, Mozart's rate of recall and composition, these are not the population mean pushed upward. They are distributions that barely overlap with the population distribution. Managing them toward the average does not improve performance. It destroys the trait you were trying to develop.
The more insidious version: people internalize this. A kid who scores inconsistently, brilliant in one context, lost in another, gets labeled unstable rather than context-dependent. An employee who produces almost nothing for six weeks and then delivers something extraordinary gets a low performance rating because the rating system measures throughput, not impact. The brilliant and the erratic look the same in the average, and the system treats them the same.
The fix is not to eliminate measurement. Measurement is not the problem. The problem is treating central tendency as representative when the question is individual.
What replaces it: measure variance, not just mean. Design for the range, not the center. The adjustable cockpit the Air Force eventually built, individual fitment, modular controls, did not just stop the crashes. It improved performance across the board, because now the cockpit served actual pilots instead of a statistical ghost.
Practically: the next time you look at an average and feel you understand something, ask what the distribution looks like. Ask who is being erased by that number. The answer is usually the people you cannot afford to lose, the outliers at the high end being flattened into conformity, and the outliers at the low end who need a different approach, not more pressure toward a standard that was never built for them.
The average is a compression artifact. Useful for certain calculations, actively harmful when used as a target.
Build for the distribution. Design for the extremes. The exceptional live there.