Wharton Human-AI Research and GBK Collective report that 80% of enterprise leaders see positive returns from AI, but only 28% call those returns strongly positive, and only 7% are at organizations that clear all three of the study's success markers. The study, released October 13 2026, defines success before counting it, according to its authors. It is still a survey, so every return in it is a person's own report.
Two things are true at once. The study is a useful map of where enterprise AI programs stand, and it shows where the evidence is thinnest.
What did the 2026 Wharton-GBK AI adoption study find?
The report, "The Adaptive Advantage: Who's Winning with AI and How," is the fourth annual study from Wharton Human-AI Research (WHAIR) and GBK Collective. It is an online survey of U.S. enterprise organizations with at least 1,000 employees and $50M in revenue, fielded July 9 to 29 2026. The sample is 1,416 respondents: 1,007 managers and above, and 409 entry- and mid-level employees. We read the release and GBK's methodology note; the full report was not available to us.
The funnel from "positive" to "all three markers"
| Measure | Share | What it measures |
|---|---|---|
| Positive returns from AI investments | 80% | Leaders reporting any positive ROI |
| Strongly positive returns | 28% | Leaders calling those returns strongly positive |
| Successful workflow transformation | 20% | Meeting all three workflow conditions |
| All three success markers | 7% | Proving value, transforming workflows, and aligning the organization |
The units shift down the table. The release describes the 20% and 7% figures as shares of organizations, and GBK's note describes 7% as the share of leaders whose organizations clear all three markers. The rows are separate measures with different definitions, not nested subsets. We cite each as published.
How the study defines the three markers
Proving value requires strongly positive ROI from a department that formally measures it. Workflow transformation requires AI fully embedded in core processes, use of agents, and significantly changed workflows. Organizational alignment requires four of five factors, including adequate training and a clear long-term vision.
Does measuring growth outcomes lead to stronger AI returns?
Leaders at organizations measuring AI against growth outcomes were more likely to report strong ROI than those focused on efficiency alone, 35% vs 13%. At the same time, 52% of leaders say AI's primary job over the next 12 to 24 months is improving productivity or cutting costs, and 22% point to growth through new revenue, products, or customer experience.
The limit matters here. This compares organizations, and, as far as the release shows, both the way an organization measures and the return it reports are self-reported by the same respondents. It does not show that a given company would raise its returns by changing its metric. GBK flags a similar confound elsewhere in the data: leaders very confident in tracking total AI spend report strongly positive returns at 56%, against 10% of less confident leaders, and the note says part of that gap may reflect conviction and not capability.
How far apart are leaders and employees on AI training and workflow change?
The gap is wide on nearly every readiness measure in the release.
| Statement | Leaders | Entry- and mid-level employees |
|---|---|---|
| Workflows have significantly changed because of AI | 73% | 58% |
| Employees receive adequate AI training | 78% | 56% |
| Employees are confident using AI in daily work | 79% | 62% |
| Leadership has articulated a clear long-term AI vision | 79% | 64% |
| AI has changed the department's management structure | 52% | 36% |
The study does not read this as employees rejecting the technology: 73% of employees say AI has had a positive effect on their own role. GBK reads the gap as being about the support around the tool. It echoes performative AI use, where what people report about AI and what they do can diverge.
What did the study find about agents that need approval for every step?
67% of leaders say their departments use AI agents in some form, but only 28% say AI is fully embedded in core processes and 17% let agents act without a person approving each step. In 48% of departments, AI performs the work and a person approves it before anything happens.
| Oversight model | Workflow transformation |
|---|---|
| Every action approved before it happens | 16% |
| A sample reviewed afterward | 18% |
| AI only advises | 25% |
| Real-time monitoring and guardrails | 31% |
GBK says the pattern holds after controlling for department, industry, company size, and seniority, and that the guardrails group is small, 68 respondents, so that figure is directional. These are department-level self-reports, so the table describes how respondents characterize their setup, not how review time or error rates were measured.
What can a survey not tell you about your own organization?
Three things. First, a survey reports what respondents believe, and GBK is direct about the weakness: self-reported ROI is a generous measure, since many respondents grade programs they might own. Second, an average across organizations does not locate your teams. Third, when leaders and employees disagree, a second survey cannot say who is right. Observed work can.
We made a related point about spend in AI cost governance: controls and dashboards show that a program exists, and they do not show that work changed.
How can an organization check its own AI returns instead of asking for them?
The study's own design suggests the method. Define the outcome first, ask more than one perspective, and name the alternative explanations. Applied to an internal rollout:
- Define strong before measuring. Write down what a strongly positive return means for each function, as one number or several, before anyone declares success.
- Set a unit of output per team. Cycle time, throughput, quality, or revenue per head, chosen before the rollout so the result is not picked afterward.
- Compare with comparable teams over the same window. Teams that adopted against comparable teams that have not adopted yet, controlling for tenure, role, and tool stack, is a stronger design than a company-wide before-and-after.
- Report confidence and the limit. A result needs a confidence level and a statement of what it cannot show, such as early adopters differing from late adopters in ways the controls miss.
| What the survey gives you | What an internal test adds |
|---|---|
| A leader's rating of returns | Output measured against a dated baseline |
| Perception of training and change | Observed use across teams, by function |
| An average across organizations | Your teams, compared with each other |
| Directional findings with caveats | A confidence score and a stated limit |
For a longer treatment of the comparison design, see how to measure AI ROI across the workforce.
Where does Levos fit?
Levos is a Human Capital Operating System: an intelligence layer above the existing stack. Surveys, governance platforms, and spend tools keep doing their jobs. Levos reads behavioral signals from the tools where work happens, M365, GitHub, Salesforce, and AI tools among them, and shows, with a confidence score, how adopting teams compare with non-adopting ones. AI Impact is one of its six signal families. The measurement methodology uses controlled cohort analysis with confidence scoring, compares adopting teams with non-adopting teams, and discloses its limits. We do not claim full attribution.
Individual-level data is visible to the person and their direct manager. Above that level, people see team aggregates of at least 5 people, apart from logged escalations of urgent issues; see the privacy model.
GBK sells research and analytics, and Levos sells measurement. Each of us has a view on what these figures should prompt, so we cite the study as published.
Frequently asked questions
What did the 2026 Wharton-GBK AI adoption study find?
Wharton Human-AI Research and GBK Collective surveyed 1,416 people at U.S. enterprise organizations in July 2026, 1,007 managers and above and 409 entry- and mid-level employees. 80% of leaders reported positive returns from AI, 28% reported strongly positive returns, and 7% were at organizations that cleared all three of the study's success markers, which are proving value, transforming workflows, and aligning the organization.
Does the study show that measuring growth outcomes raises AI returns?
No. Leaders at organizations that measure AI against growth outcomes were more likely to report strong ROI than those focused on efficiency alone, 35% vs 13%, but this compares organizations, and, as far as the release shows, both the measurement approach and the return are self-reported. The study does not show that switching metrics would raise a given organization's returns.
How far apart are leaders and employees on AI training and workflow change?
78% of leaders believe employees receive adequate AI training, against 56% of entry- and mid-level employees. 73% of leaders say workflows have significantly changed because of AI, against 58% of employees. Both are survey answers, so neither is an observation of what changed.
What did the study find about agents that need approval for every step?
In 48% of departments, AI performs the work and a person approves it before anything happens. Workflow transformation was 16% where every action is approved and 18% where a sample is reviewed afterward, against 31% under real-time monitoring and guardrails and 25% where AI only advises. The guardrails group is 68 respondents, so GBK says to read that figure as directional.
How can an organization check its own AI returns instead of asking for them?
Define what strong means before measuring. Set a unit of output for each team, take a dated baseline, and compare teams that adopted with comparable teams that have not adopted yet over the same window. Report a confidence level and a stated limit alongside the result.
See what changed in your own teams
If your evidence of AI returns is a survey or a leader's rating, it can be checked against observed work. Request a Demo and we will walk through what a comparison-based readout contains, or read how the measurement methodology works first.
Levos is accepting design partner applications from US organizations of 100 employees or more with an active AI rollout. Large organizations typically start with one function or division, measured against comparable teams that have not adopted yet. That is a stronger attribution design than a company-wide before-and-after.