AI boosts the person before it boosts the company. McKinsey's August 25, 2026 survey found 80% of respondents report individual productivity gains but only 37% see enterprise profit impact, and an Atlanta Fed study of 6,000 executives measured just 0.29% firm-level gain over three years. Daugherty and Wilson's Radically Human names the fix: leapfrog companies redesign workflows around the technology instead of bolting it on.
Monday brought two datasets that put hard numbers on the most common complaint I hear from founders about AI: everyone on the team swears they are faster, and the P&L cannot find any of it. On August 25, 2026, McKinsey published its survey The State of AI in 2026: On the Road to ROI, and coverage landed the same day of an Atlanta Federal Reserve study of senior executives across four economies. They describe the same gap from two altitudes.
The Atlanta Fed team surveyed close to 6,000 senior executives in the US, UK, Germany, and Australia between November 2025 and January 2026. Their headline: 89% of firms reported no impact from AI on labor productivity over the past three years, and when the reported effects were averaged out, AI had lifted firm productivity by roughly 0.29%. This is not a story of non-adoption — 69% of the surveyed businesses already use at least one AI technology, with US adoption at 78%. McKinsey's survey of 1,719 respondents across 97 countries, fielded May 4 to June 8, 2026, supplies the other half: 80% of respondents say AI improves their individual productivity, yet only 37% report any enterprise-level EBIT impact — a share that is flat against last year's survey.
| What gets measured | The number | Source, published Aug 25, 2026 |
|---|---|---|
| Individuals reporting AI productivity gains | 80% | McKinsey State of AI 2026 |
| Companies reporting profit (EBIT) impact | 37%, flat year over year | McKinsey State of AI 2026 |
| Firms reporting no labor-productivity impact in 3 years | 89% | Atlanta Fed executive survey |
| Average firm-level productivity gain, 3 years | 0.29% | Atlanta Fed executive survey |
| Respondents who can tie AI to significant value | ~6% ("high performers") | McKinsey State of AI 2026 |
The gap is not a paradox once you look at what each number measures. Individual gains are task-level: a faster draft, a quicker analysis, a meeting summary you did not write. Firm productivity is workflow-level, and a workflow is only as fast as its slowest handoff. If your associate produces the memo in an hour instead of four but it still waits two days for review, sits in the same approval chain, and feeds the same meeting, the company has captured nothing. The saved time leaks into more polish, more Slack, or simply earlier log-off — none of which shows up in EBIT. McKinsey's own data says the roughly 6% who do capture value share one trait: they fundamentally redesigned workflows rather than adding AI to existing ones.
This is precisely the argument Paul Daugherty and H. James Wilson made in Radically Human, and the new numbers read like a delayed confirmation. Their Accenture research found that most companies adopt technology as a lifeline — patching existing processes — while a minority of "leapfroggers," about 18% in their study, broke previous performance barriers by rebuilding processes around what the technology makes possible and grew roughly four times faster than laggards. The lifeline pattern is exactly what an 89%-no-impact statistic looks like at scale: widespread adoption, untouched process architecture.
There is also a human variable that founders underrate. University of Pittsburgh professor Mark Ma, commenting on the related NBER data, notes that employee sentiment toward AI is one of the strongest predictors of firm-level productivity from AI — and that companies justifying layoffs with AI are actively destroying the conditions the gains depend on, with stock-market reactions to those announcements averaging near zero. Meanwhile the Atlanta Fed found executives personally average about 1.5 hours of AI use per week. Leaders are mandating transformation on tools they barely touch, then wondering why the culture will not carry it.
What I actually do with clients now: pick one workflow with a real cycle time — proposal creation, candidate screening, weekly reporting — and redesign it end to end, removing the steps AI makes unnecessary, before touching a second workflow. Measure cycle time before and after, not sentiment. One rewired workflow that closes in two days instead of seven is worth more than forty licenses of anything.
The honest limits: both studies are self-reported surveys, and executives are bad estimators of their own firms' productivity. Three years is early — electricity took decades to show up in productivity statistics, and the surveyed executives themselves expect a 1.4% gain over the next three years, 2.3% among US firms. The 0.29% average also hides a distribution; some firms are quietly compounding real advantages. But if you are betting your company on AI this quarter, bet on redesign, not on adoption. Adoption is the part that is already priced at zero.
Sources: explainx.ai's August 25, 2026 breakdown of McKinsey's The State of AI in 2026 (explainx.ai/blog/mckinsey-state-of-ai-2026-roi-agentic-coding-august-2026); Business Standard's coverage of the Atlanta Fed executive survey (business-standard.com); Futurism on the NBER working paper and Mark Ma's analysis (futurism.com); Paul Daugherty and H. James Wilson, Radically Human (Harvard Business Review Press, 2022).
Related: How to Find Your Passion · Best Self-Improvement Books · How to Make Better Decisions · What University Will Not Teach You