---
type: "synthesis"
clusters: ["C6", "C4"]
id: "meta-measurement-problem"
sources: ["cross-day"]
---
The corpus is strong on mechanism, weak on magnitude, and unanimous that traditional ROI mismeasures AI. Spine: [[claim-traditional-roi-fails-ai]], the [[prereq-productivity-j-curve|J-curve]] dip makes short-term numbers look bad, and each article proposes a replacement ([[cd-roi-is-the-wrong-lens]]). Execution: GenAI is the hardest-to-value technology ([[claim-genai-hardest-to-value]], [[cross-genai-measurement-problem]]); payback is compressing ([[concept-compressed-ai-payback]]). Adoption: everyone rejects vanity metrics but no shared replacement exists ([[cross-measurement-problem]]); reward *documented judgment*, not raw output ([[cross-incentives-metrics-redesign]]). Reskilling: cost-per-learner and training hours measure exposure, not capability ([[cross-completion-not-capability]]). Tail: the metric is not the reality ([[cross-measurement-artifact-vs-reality]], [[cross-metrics-mislead-managers]]). Practical rule: pair every easy metric with a truer lagging signal, and value proprietary data + workflow over generic prose. Companion: [[meta-task-to-process-gap]].