Accountable Acceleration: Wharton/GBK 2025 AI Adoption Report
TL;DR
The third wave of an annual survey of US enterprise decision makers by Wharton Human-AI Research and GBK Collective. About 800 people answered between 26 June and 11 July 2025 (PDF p. 6). Its 74% is often set against MIT NANDA’s 95%.
- Three in four report a positive return. 74% of all respondents say their organisation’s GenAI ROI to date is positive: 35% “significantly positive” and 39% “moderately positive” (PDF p. 47; text, p. 38). The base is all respondents (n=801), not only firms that measure ROI.
- Most say they measure it. 72% say their organisation measures GenAI ROI in a fully integrated (37%) or consistent, standardised (35%) way (PDF p. 45; p. 7).
- Use is now routine. 82% use GenAI at least weekly (+10 points on 2024) and 46% daily (+17 points) (PDF p. 7).
- The largest firms are least sure. In Tier 1 firms ($2B+ revenue), 57% report positive ROI and 25% say it is “too early to tell/still in pilot phase” (PDF p. 47).
- Seniority shapes the answer. 81% of VP-and-above see positive ROI against 69% of managers and directors (PDF p. 12, p. 48).
- People are the stated bottleneck. 89% agree GenAI enhances employees’ skills, and 43% see a risk of declining skill proficiency (PDF p. 7).
Key data
| Figure | Definition | n | PDF page |
|---|---|---|---|
| 74% positive ROI | 35% significantly + 39% moderately positive; answer to “Based on internal conversations with colleagues and senior leadership, what has been the return on investment (ROI) from your organization’s Gen AI initiatives to date?“ | 801 | 47 |
| 9% / 3% / 12% | neutral / negative / too early to tell | 801 | 47 |
| 72% measure ROI | ”fully integrated, strategic and continually optimized” (37%) + “measured consistently using standardized methods” (35%) | 801 | 45 |
| 57% / 25% | Tier 1 ($2B+): positive ROI / too early to tell | 155 | 47 |
| 81% vs 69% | positive ROI, VP/C-suite/owner vs manager/director | 372 / 429 | 48 |
| 88% / 83% / 75% / 54% | positive ROI in Tech/Telecom, Banking/Finance and Professional Services, Manufacturing, Retail | 107–197 per industry | 49 |
| ~4 in 5 | expect positive ROI within two to three years (Tier 1: 71%) | 801 | 50 |
| 82% / 46% | use GenAI at least weekly / daily | ~800 | 7 |
| 16% | “lagging” decision makers who use GenAI less than weekly | ~800 | 10 |
| 60% | organisations with a Chief AI Officer or similar role (26% new position, 34% added to an existing role) | 801 | 68 |
| ~30% | share of GenAI technology budgets going to internal R&D (IT respondents) | n/a | 12 |
Method and definitions
- Instrument (PDF p. 6): a 15-minute online quantitative tracking survey, United States only. “Interviews conducted between June 26 and July 11, 2025.” Respondents: ~800 in 2025 (n=801 on the charts), ~800 in 2024, ~670 in 2023.
- Who answered. Senior decision makers in HR, IT, Legal, Marketing/Sales, Operations, Product/Engineering, Purchasing/Procurement, Finance/Accounting or General Management. Their employers are US commercial organisations with 1,000+ employees and more than $50 million in revenue (PDF p. 6). Charts split them into Tier 3 ($50M–<$250M, n=218), Tier 2 ($250M–<$2B, n=428) and Tier 1 ($2B+, n=155).
- Unit. People, not firms. Each answer is one decision maker’s view of their organisation. There is no firm-level deduplication or financial data.
- What counts as a “positive return”. A perception gathered through “internal conversations with colleagues and senior leadership”. “Moderately positive” is defined as “measurable benefits, but limited or in early stages”, which is a low bar. The question is new in 2025, so there is no earlier ROI figure to compare (PDF p. 47).
- Repeated cross-section. Questions and answer options changed between waves, and “certain questions cannot be tracked across each year” (PDF p. 88).
- Sponsor and fieldwork. The study “is sponsored by Wharton Human-AI Research …; GBK Collective performed research and analysis”. GBK led the 2023 wave with Puntoni, and the joint study began in 2024 (PDF p. 4). GBK describes itself as a “marketing strategy and insights consultancy” (PDF p. 4). Korst, a GBK partner, is also “founder of Mindspan Labs—a forward-thinking AI consultancy and incubator” (PDF p. 3).
Scope and reliability
- Self-report by optimistic respondents. Every figure is a decision maker’s own judgement. The report shows that seniority moves the answer: VP-and-above are twice as likely as managers to say their organisation adopts “much faster” than others (56% vs 28%, PDF p. 10).
- Generous ROI definition. Puntoni told Big Technology: “Our definition is very different from the MIT report. Theirs is much more stringent.” Big Technology adds that the Wharton study counted individual productivity, employee retention and operational efficiency (Kantrowitz, 31 Oct 2025; raw at
raw/articles/2025-10-31-kantrowitz-big-technology-are-74-percent-of-businesses-seeing-genai-roi.md). - Base. Big Technology wrote that 74% “of businesses that measure the ROI” see a positive return. The chart’s base is all respondents (“Among Total”, n=801). The executive summary’s sentence “Most firms now measure ROI, and roughly three in four already see positive returns” (PDF p. 8) invites the narrower reading.
- Sample frame. US only. “Enterprise” starts at $50 million in revenue, so more than a quarter of respondents work at firms below $250 million.
- Who paid. Big Technology reports that “Wharton paid for the study” and describes GBK as “a company that does some AI advising”. The report itself names no commercial sponsor.
Dynamic-capabilities reading
digital-sensing/digital-scouting: a yearly scan of how US enterprises take up GenAI, by function, industry, firm size and seniority.digital-transforming/improving-digital-maturity: the human-capital section measures skills (89% enhance, 43% atrophy risk), training budgets, hiring and Chief AI Officer roles (PDF pp. 58–81).contextual/internal-barriers: the top barriers are security risk, operational complexity and data inaccuracy. Lack of training resources enters the top ten. “Laggards” cite employee resistance and lack of trust more often (PDF pp. 39, 57).
How it connects
- enterprise-ai-adoption: a broad, self-reported value figure (74%) from US enterprises, beside adoption figures of 82% weekly use.
- ai-maturity-measurement-comparison: a “broad definition” instrument. It counts a moderate, early benefit as a positive return.
- MIT NANDA’s GenAI Divide: the stricter 2025 instrument this report is usually set against, as Big Technology did.
- BCG 2025: a global executive survey from the same season that scores capabilities rather than asking about ROI directly: 5% future-built, 35% scaling.
- micro-productivity-trap: the report’s top benefit is employee productivity for the third year running (PDF p. 39). Its ROI question does not separate task-level gains from firm-level results.
- ai-deskilling: 43% of leaders see a risk of declining skill proficiency (PDF p. 7).
What was actually ingested
The 90-page Full Report PDF from ai.wharton.upenn.edu, converted with pdftotext -layout. The ROI and ROI-measurement charts (PDF pp. 45, 47) were read from page renders, because conversion loses the legend. The Knowledge at Wharton landing page refused curl (HTTP 403) and was read with WebFetch for its date (28 October 2025) and PDF links. The 14-page Executive Summary repeats PDF pp. 6–14 and was not kept. Page numbers here are PDF pages; the printed slide number is two lower.
Linked entities and concepts
- enterprise-ai-adoption, micro-productivity-trap, ai-deskilling: concepts.
- ai-maturity-measurement-comparison: thread.
Dangling (single-source mention, deferred): Jeremy Korst (GBK Collective partner; Mindspan Labs founder), Stefano Puntoni and Prasanna Tambe (faculty co-directors, Wharton Human-AI Research), GBK Collective, Wharton Human-AI Research / The Wharton School, Alex Kantrowitz (Big Technology), Aaron Levie (Box CEO, quoted by Big Technology).