Four years into the Wharton Human-AI Research and GBK Collective tracking study, artificial intelligence adoption is no longer the story.
Among managers and above, 62 percent now use AI daily, and 73 percent say it has significantly changed how their department works. Positive returns are near-universal, at 80 percent.
The problem sits one rung higher.
Only 28 percent report strongly positive returns, and just 7 percent of organizations succeed on all three of the study’s success markers: proving value, transforming workflows, and aligning the organization behind a cohesive strategy.
That 52-point gap between positive and strongly positive is where executive attention now belongs.
Funding for Transformation, Measuring for Convenience
Here is the tension at the center of the data. Some 83 percent of managers and above expect Applied-AI to fundamentally transform or majorly affect their industry within two to five years.
Yet the most common test departments apply to their own AI spend is hours saved (40 percent), while net-new revenue is the least used of the major measures (27 percent).
Asked what AI’s primary job should be over the next 12 to 24 months, 52 percent name productivity or cost reduction, and only 22 percent name growth.
The returns follow the yardstick. Departments that track at least one growth measure, such as net-new revenue, sales conversion, output volume, or customer satisfaction, report strongly positive returns 35 percent of the time.
Departments judging AI on efficiency alone report 13 percent. In my view, hours saved is the easiest return to claim and the hardest to convert into something a board recognizes as value creation. A leadership team that budgets for transformation and accepts time-sheet math as proof has set the scoreboard to the wrong game.
Spending visibility compounds the problem. Only 40 percent of managers and above are very confident their organization can track total AI spend once costs are bundled into other software and cloud services are counted.
Among that confident group, 56 percent report strongly positive returns, against 10 percent of the less confident. The study’s authors caution that confident respondents are also more enthusiastic about AI, so part of that gap may reflect conviction rather than capability.
Even so, 87 percent expect budgets to rise over the next 12 months, and the cost of tools, compute, and licenses is already the most-cited barrier to stronger returns (34 percent). No CFO can defend a budget line the organization cannot fully see.
The Supervised Middle Is the Costly Place to Live
Two-thirds of departments (67 percent) use AI agents in some form, yet only 17 percent permit genuine autonomy. In 35 percent of departments, AI drafts and a person makes every final call. In the largest group, 48 percent, AI does the work and a person approves it before anything happens.
The success pattern is U-shaped. Workflow transformation succeeds in 31 percent of departments using Agentic AI to its fullest capacity and in 25 percent where AI only advises. It falls to 16 percent where every action is approved and 18 percent where a sample is checked afterward.
Returns follow the same shape. In my reading, organizations stalled mid-journey are paying for agents and reviewers at the same time, and collecting the full reward of neither.
Accountability is the unfinished business. No single owner of agent errors reaches 30 percent: governance or risk teams lead at 28 percent, the deploying employee’s manager follows at 26 percent, the employee at 25 percent, and the vendor at 9 percent.
Another 11 percent say nobody is clearly accountable or that they cannot say, and that uncertainty peaks in the middle of the hierarchy, at 20 percent of managers and directors against 3 percent of VP-level and above.
Autonomy tracks risk tolerance more than technical capability. Agents may complete low-impact internal tasks without sign-off in 66 percent of departments, routine operations in 63 percent, actions that touch customers in 45 percent, and irreversible financial or legal actions in 32 percent.
Notably, 42 percent of C-suite respondents say their department lets AI take irreversible actions, against 24 percent of managers and directors. The people who set policy and the people who run the work see autonomy differently.
The Seniority Gap Is the Most Consistent Finding
Only 39 percent of managers and above describe an environment that clears at least four of the study’s five alignment factors. Among entry- and mid-level employees, the figure is 26 percent.
The gap appears in every industry and company size, and the study found no group where managers and the people they manage describe the same working environment.
Training shows the widest divide: 78 percent of managers say employees receive adequate AI training, while 56 percent of entry- and mid-level employees agree. A clear long-term AI vision registers at 79 percent versus 64 percent.
Marketing and Sales, the functions charged with telling the external AI story, are the least likely to report internal alignment (25 percent).
User sentiment is cooling even as usage climbs. Among managers and above, “excited” fell 8 points to 51 percent, “cautious” rose 7 points to 44 percent, and “skeptical” rose 8 points to 25 percent.
This is the first wave in which usage and sentiment moved in opposite directions. The workforce is not hostile, since 68 percent of entry- and mid-level employees say their view of AI improved over the past year.
It is reluctant, and still being carried along.
Workforce decisions matter here. Departments with strong returns did not cut more staff or freeze more hiring than others. They reorganized instead.
They were more likely to have changed management structure (64 percent versus 47 percent), monitor AI use (92 percent versus 74 percent), include it in performance reviews (87 percent versus 65 percent), and train employees adequately (91 percent versus 73 percent).
Where departments did cut staff because of AI, only 20 percent of managers and above describe an aligned organization, against 53 percent where no cuts occurred. Meanwhile, 82 percent say AI helps newer employees contribute faster, yet 24 percent expect fewer entry-level hires over the next two to five years, versus 14 percent for executive roles.
Leaders thinning that talent pipeline should be clear about where tomorrow’s experienced managers will come from.
Governance Choices That Travel With Returns
Organizations with a newly designated Chief AI Officer report strongly positive returns 58 percent of the time. That drops to 24 percent where the duty was added to an existing executive’s role, and 8 percent where no such role exists or is planned.
Two-thirds of organizations now report a CAIO or equivalent, but nearly all of this year’s growth is reassignment rather than dedicated appointment.
Platform posture shows a similar pattern.
Strongly positive returns reach 50 percent among organizations standardized on a single approved platform, against 30 percent for those allowing multiple approved tools under a formal policy, 19 percent for those running a central hub or orchestration layer, and 9 percent for those still experimenting without a defined approach.
As an advocate of open choice, I read that last number as the real warning.
The data rewards decisiveness. It cannot tell us whether the single-platform premium reflects simpler governance or something closer to lock-in. Notably, 39 percent of organizations deliberately mix proprietary and open-weight models, while 30 percent run mostly proprietary and 28 percent mostly open-weight.
Outlook for AI Strategic Growth Applications
First, rewrite the scorecard. Every material Applied-AI Initiative should carry at least one growth measure alongside its efficiency metrics, whether net-new revenue, conversion, output volume, or customer satisfaction.
If a business case cannot name one, that's a signal about the business outcome.
Second, build the spend ledger before the next budget cycle. With 87 percent of leaders expecting higher budgets, finance will be asked to defend numbers that bundled licenses and embedded features currently blur.
Third, choose a destination on the autonomy ladder and tier it by risk. Low-impact internal tasks can move toward real autonomy with monitoring and guardrails, while irreversible financial and legal actions can keep human sign-off by design. What leaders should avoid is drifting in the middle indefinitely.
Name an accountable owner for agent errors at the same time.
Fourth, treat alignment as an ROI lever rather than a human resources initiative. A 13-point gap on strategy alignment and a 22-point gap on training are execution problems with a measurable cost. The workforce is not resisting. It is waiting to be told where the organization is going, and to be equipped to get there.
Adoption was the easy part, and every competitor has now done it. The question for your next board meeting is not how many employees use AI, but which growth number moved because they did.
Reach out to learn more about our Applied-AI Initiative objectives.