Wharton–GBK《The Adaptive Advantage: Who’s Winning with AI and How》報告揭示企業 AI 投資回報差距

Wharton Human-AI Research and GBK Collective’s new report, The Adaptive Advantage: Who’s Winning with AI and How, shows how far enterprise AI has spread—and how much harder it is to turn adoption into substantial business returns. Eighty percent of managers and more senior respondents report positive returns on their department’s AI investments, but only 28% describe those returns as strongly positive. The fourth annual study points to workflow redesign, clearer ownership, and growth-oriented measurement as distinguishing features of organizations reporting stronger results.

Released on October 13, the study moves the enterprise AI discussion beyond whether employees use the technology. Its more consequential question is whether companies have changed the processes, incentives, and management structures around it.

Measuring Growth Changes the Picture of AI Value

One of the report’s sharpest contrasts concerns what departments count as success. Among those tracking at least one growth measure, 35% report strongly positive AI returns, compared with 13% among those measuring efficiency alone. That makes strong returns roughly 2.7 times as common in the growth-measurement group.

The distinction is broader than revenue. The study classifies net-new revenue, sales conversion, output volume, and customer satisfaction as growth measures. Its efficiency measures cover operational costs, hours saved, turnaround times, and error rates. Departments could select up to three indicators, so the growth group can include organizations that also track efficiency.

This matters because time savings remain easier to describe than business expansion. Hours saved is the most commonly selected measure, at 40%, while net-new revenue is selected by 27%. Meanwhile, 83% of managers and above expect AI to fundamentally transform or substantially affect their industry within two to five years.

The practical implication is to connect AI projects to the outcomes the business actually wants. A faster sales workflow, for example, can be evaluated through both time saved and conversion performance. The survey does not establish that changing a dashboard causes better returns; it shows an association between a broader measurement approach and stronger reported results.

AI Agents Are Common, Deep Integration Is Scarce

Agent adoption is already widespread: 67% of departments report using AI agents in some form. Yet only 28% describe AI as fully embedded in their core business processes, and just 17% permit autonomous operation with human review occurring afterward or replaced by monitoring and guardrails.

The largest group, 48%, says AI performs work but a person reviews and approves it before action. Another 35% uses AI for suggestions or drafts while humans make final decisions. Those distinctions show why an “agent deployed” milestone says little about how much a workflow has actually changed.

The report’s workflow-transformation marker requires three conditions together: significantly changed workflows, AI embedded in core processes, and agent use in some form. Among departments meeting that combined definition, 67% report strongly positive ROI.

That finding supports a more demanding view of deployment. Installing an assistant is only one part of the work; teams also need to decide how information moves, which exceptions require intervention, and who owns the result. It does not imply that every task should be autonomous. The report identifies security, regulation, and trust as barriers to granting agents greater freedom, making the suitability of each workflow central to the decision.

Clear Ownership Accompanies Stronger Returns

Organizations with a newly designated Chief AI Officer position stand out. Among managers and above in those organizations, 58% report strongly positive returns, compared with 24% where AI oversight was added to an existing executive’s responsibilities. The figure is 8% where no such role exists and none is planned.

Platform strategy shows another divide. Strongly positive returns are reported by 50% of respondents whose organizations standardize on one approved AI platform, compared with 9% of those still experimenting without a defined approach. These comparisons are observational: dedicated leadership and platform consistency may accompany other advantages, including resources and organizational maturity.

Accountability for agent errors is less settled. Respondents assign responsibility most often to governance or risk teams (28%), the deploying employee’s manager (26%), or the employee (25%). No single answer dominates. For businesses expanding agent use, defining responsibility is part of operational design, alongside permissions and monitoring.

Employees See a Different Transformation

For the first time, the annual study includes entry- and mid-level employees alongside managers and senior leaders. The comparison exposes a substantial difference in how readiness is perceived.

While 73% of managers and above say departmental workflows have changed significantly because of AI, 58% of entry- and mid-level employees say the same. The training gap is larger: 78% of leaders believe employees receive adequate AI training, compared with 56% of employees.

These differences do not amount to wholesale employee rejection. Seventy-three percent of entry- and mid-level employees say AI has positively affected their role, and 84% of those using it say it saves time. The tension is between individual usefulness and the support needed to make adoption work across an organization.

Workforce consequences also remain unsettled. Thirty-seven percent of managers and above report AI-related staff reductions, and 33% report slowed or frozen hiring. However, departments reporting strong returns are no more likely than others to report those cuts or hiring freezes. Stronger returns instead accompany changes in management structures, workflows, monitoring, and training.

What the Survey Can—and Cannot—Tell Businesses

The research surveyed 1,416 full-time employees at U.S. private-sector enterprises with at least 1,000 employees and more than $50 million in revenue. Interviews took place July 9–29, 2026. The sample comprises 1,007 managers and above and 409 entry- and mid-level employees, with the two audiences reported separately.

These are self-reported assessments, rather than independently audited investment returns. The results describe large U.S. commercial enterprises and should not automatically be generalized to small businesses or other countries. Comparisons with earlier years use the managers-and-above population, since junior employees were newly included in 2026.

For enterprise teams, the report offers a useful test of AI progress: look beyond usage counts to measurable business outcomes, integration into core work, clear accountability, and employees’ actual experience. Widespread access can coexist with weak execution, and reported productivity improvements can leave substantial organizational work unfinished.

Readers who wish to learn more about the findings, methodology, and industry and departmental comparisons can read the full Wharton–GBK report, The Adaptive Advantage: Who’s Winning with AI and How.