---
type: "synthesis"
tags: ["synthesis", "measurement", "roi"]
sources: ["execution"]
id: "cross-genai-measurement-problem"
sourceVaultSlug: "hbr-seg-execution"
originDay: 8
articleStem: "hbr-seg-execution"
sourceUrl: "(unified vault: 7 sources)"
sourceTitle: "HBR — Firm Ⅱ-C · Execution quality — correct execution of AI"
---
## Measurement difficulty is the enabler of every other failure

A quiet through-line: firms act on AI value they cannot actually measure.

- **A062**: [[claim-genai-hardest-to-value]] — 44% of executives call generative AI the *hardest* AI to value economically ([[contrarian-genai-hardest-to-value]]), harder than analytical, deterministic, or agentic AI. [[concept-ai-economic-value-measurement]] is the enabler of [[concept-anticipatory-ai-layoffs]] — you can't justify displacement by a value you can't quantify, so firms act on faith.
- **A060**: [[question-defining-ai-roi]] — the article contrasts activity metrics with business outcomes but never publishes the KPIs; [[concept-pilot-theater]] thrives where measurement is weak.
- **A089**: [[claim-converged-payback-period]] — leaders converged to a measurable 6–12 month payback precisely because they built measurement and data discipline ([[concept-compressed-ai-payback]]).
- **A077**: [[claim-marginal-business-impact]] — measured through one social-listening corpus, impact reads as marginal.

## The synthesis

Measurement is the hinge. When it's absent, you get pilot theater (A060), anticipatory layoffs (A062), and hype-driven expectations ([[contrarian-ai-hype-vs-reality]], A077). When it's present, you get converged payback and disciplined scaling (A089) and the [[action-controlled-experiments|A/B experiments on narrow-deep use cases]] A062 prescribes. **The recommended cure across the corpus is the same: controlled experiments on isolated use cases before any structural commitment.** See [[cross-action-vs-inaction-paradox]].