---
type: "synthesis"
tags: ["synthesis", "roi", "performance-gap"]
sources: ["execution"]
id: "cross-winners-losers-execution-gap"
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"
---
## Failure is the base rate; the winners are separating

Four articles quantify the same landscape from different angles, and the numbers rhyme:
- **[[claim-95-percent-failure]]** (A060): 95% of GenAI programs deliver no bottom-line return.
- **[[claim-marginal-business-impact]]** (A077): real usage is 'modest, uncontroversial wins,' core processes rarely rethought — [[contrarian-ai-hype-vs-reality]].
- **[[claim-widening-performance-gap]]** (A089): AI leaders moved from 2.7x to 3.8x the bottom half via [[concept-compounding-ai-capabilities]], even as payback converged for everyone ([[concept-compressed-ai-payback]]).
- **[[entity-moodys]]** (A093): a concrete member of the winning minority.

## The tension the corpus never fully resolves

If 95% fail yet leaders compound and payback has compressed to 6–12 months, is the gap widening or closing? The reconciliation: **speed of return democratized, magnitude of advantage diverged.** The commodity ecosystem lets anyone reach fast payback *if they reach production at all* — but only disciplined executors do, and their early wins compound. [[question-laggard-catchup-viability]] stays open: A089 calls catch-up a 'distinct possibility,' yet the gap grew. The other six articles supply the missing variable — execution quality (leadership, trust, process redesign, measurement) is what separates the 5% from the 95%. See [[cross-the-execution-quality-thesis]] and [[cross-leadership-differentiator]].