The Average-Fit Loss: A Unifying Metric for the Cost of One-Size-Fits-All Systems

Products, models, and policies are routinely optimized for a representative or average member of the population they serve. We define the Average-Fit Loss (AFL), a population-level functional that quantifies the utility forgone when a single design tuned to the center of a heterogeneous population is applied to that population in full, and we use it to bring several established results from separate literatures under one lens. Under a smooth fit model the AFL equals a dispersion functional of the population, recovering the mismatch cost of address models of product differentiation; in the quadratic case it is the total variance. Once participation is endogenous, the AFL decomposes into an intensive margin (degraded service to those retained) and an extensive margin (exclusion of those pushed below threshold), the latter being the deadweight loss of unserved demand. Because the excluded do not appear in a system's own usage data, any AFL estimate formed from observed users is downward biased, an instance of selection and survivorship bias applied to the measurement of fit: a system cannot, from its own data, see most of its own average-fit loss, and the bias grows with the mass excluded. In more than a few dimensions almost no one is near-average on every relevant dimension, so the loss is borne broadly rather than by a minority. We illustrate these on the ANSUR II anthropometric survey (6,068 individuals, 93 body measurements), the modern descendant of the 1952 cockpit data. The contribution is not a new theorem but a single named metric, and a measurement, that connect product differentiation, welfare economics, and selection bias to the design of systems built for the average. The AFL is the formal complement of the price of fairness: not the cost of imposing accommodation, but the cost of withholding it. A single self-contained script reproduces every empirical number and both figures in the paper.