---
type: "synthesis"
tags: ["incentives", "performance-management", "accountability", "recurring-lever"]
articles: ["a036", "a037", "a040", "a041"]
id: "cross-incentives-metrics-redesign"
sources: ["adoption"]
sourceVaultSlug: "hbr-seg-adoption"
originDay: 9
articleStem: "hbr-seg-adoption"
sourceUrl: "(unified vault: 11 sources)"
sourceTitle: "HBR — People Ⅲ-A · Adoption / trust / literacy / psych-safety"
---
A recurring, under-appreciated lever: employees' resistance to AI is often a *rational response to broken incentives*, so the fix is redesigning what gets rewarded — not exhortation.

- **A036** — [[claim-input-metrics-punish-efficiency]]: measuring hours/effort punishes the worker who finishes 40% faster with AI, so they hide it. Fix: reward output ([[action-reward-output-over-input]]), even reward working *less* time ([[contrarian-rewarding-less-work]]).
- **A041 (Pernod Ricard)** — the sharpest formalization: [[concept-span-of-control-vs-accountability]]. AI shrinks a worker's *control* but organizations rarely shrink their *accountability*, creating a negative incentive. The fix is the safe harbor ([[concept-risk-free-adoption]], [[action-restructure-evaluations]]) captured in [[quote-safe-harbor-compliance]] — follow the AI and miss quota = OK; ignore it and miss = not OK.
- **A040** — [[contrarian-metric-penalties]]: traditional frontline metrics (time-clocks, scan rates) *punish the trial-and-error AI requires*; experimentation feels unsafe.
- **A037** — [[claim-financial-incentives-dampen-transparency]]: outcome-tied bonuses actively *suppress* the desire for AI explanations; fix via [[action-align-incentives-critical-engagement]].

Synthesis: four articles converge on a design principle — **align the reward system with the behavior you want (adoption, transparency, experimentation) or workers will rationally do the opposite.** Note the internal tension A037 vs A036: A036 wants pure output rewards; A037 warns pure outcome rewards breed willful blindness. The resolution is to reward *documented judgment* (override justifications, error-catching), not raw speed — a bridge to the measurement problem ([[cross-measurement-problem]]).