---
type: "synthesis"
tags: ["synthesis", "process-redesign", "productivity"]
sources: ["execution"]
id: "cross-task-to-process-translation"
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"
---
## The single most-repeated idea in the corpus

Four articles independently locate the same failure point: individual, task-level AI gains do not automatically become organizational value. This is the corpus's master mechanism.

- **A062** names it directly: [[concept-individual-vs-process-productivity]]. A measured 10–15% coding boost is real, but [[claim-translation-difficulty]] — translating it into end-to-end efficiency is 'difficult to say the least.' Employees report *smaller* gains than the C-suite expects.
- **A054** shows the dark version: [[claim-verification-negates-productivity]] — verification labor can wipe out the generation-phase savings, so sequential AI use can *decrease* net productivity ([[concept-productivity-paradox]]). The fix is [[claim-process-redesign-required]] and [[action-redesign-interorganizational-processes]].
- **A077** measures the outcome: [[claim-marginal-business-impact]] — 'core business processes are rarely rethought.'
- **A089** supplies the counter-example: leaders that invested in data management and workflow integration (the four pillars) converted gains into ~500% ROI cases.

## Why it matters

The prescriptions converge: measure narrow-deep use cases with controlled experiments ([[concept-narrow-deep-use-cases]], [[action-controlled-experiments]]), then **redesign the process** with employees ([[action-redesign-business-processes]]) rather than bolting AI onto old workflows. The recurring warning of [[cross-the-execution-quality-thesis]]: don't ask whether AI is better at a task; ask whether AI taking the task makes the *process* better. See [[cross-roi-leakage-attribution]] for where the untranslated value actually goes.