How AI changes what the measures of cost, productivity, and return actually measure, and the disciplines that rebuild them.
Firms are building their first broadly institutionalized generation of AI value measures: return on investment, cost per outcome, output per employee, spend as a share of revenue, utilization, deflection, and new consumption-based procurement benchmarks. The arithmetic of these ratios is familiar. The objects being divided increasingly are not. AI-driven change in a measurement object is not itself distortion; distortion arises when that change violates a comparability assumption required to read the measure as evidence of value, and goes unrecognized or unadjusted.
The paper sets out four mechanisms: substitution shrinking the cost and headcount bases that AI spend is judged against, from inside the measurement period; automated triage sorting the populations being averaged; tokens and tasks changing content while keeping their names; and per-seat pricing giving way to credits, actions, and conversations that no benchmark can compare. Each has precedents; their conjunction, speed, and interiority do not. It then develops the matching problem those changes feed, traces how naive reliance misallocates capital, and sets out the disciplines and a six-question ratio integrity test by which the ratios are rebuilt, with the open problems named.
This is a living draft. Versions are numbered and dated on the title line. Comments and criticism are welcome: contact@jamesemurphy.com.
James E. Murphy, “Rebuilding Ratios: AI Distortion in Value Measurement,” working draft v1.2, August 2026. Available at https://jamesemurphy.com/ai-economics/measurement/ and https://ssrn.com/abstract=7262118, doi:10.2139/ssrn.7262118.