Wharton 2026 AI Report: 80% Report Positive ROI, Only 28% Strong Returns

The fourth annual enterprise AI study from Wharton Human-AI Research and GBK Collective, published on 13 October 2026, finds that AI adoption among large US companies is close to universal and that returns are wide but shallow. Among about 1,000 managers and executives, 80% report a positive return on their AI investments to date, but only 28% report a strongly positive one, according to the full report. The study's title, "The Adaptive Advantage", carries its argument: the difference between the 28% and everyone else is how much they changed the work, not how much AI they bought.

What happened

The survey was a 15-minute online questionnaire conducted between 9 and 29 July 2026 with roughly 1,400 full-time employees of US commercial organisations with at least 1,000 staff and more than $50 million in revenue. For the first time it split respondents into "managers+" (1,007 managers, directors, VPs, C-suite and founders) and 409 entry- and mid-level employees, so it can compare what leadership reports with what the workforce sees.

Usage is now routine. Among managers+, 62% use AI daily, up 16 points, and 91% use it more than weekly. Seventy-three percent say their workflows have changed significantly because of AI, though only 28% describe AI as fully embedded in core workflows. Sentiment is cooling as use rises: "cautious" was up 7 points and "skeptical" up 8, while "excited" fell 8.

Agents are widespread on paper and constrained in practice. Sixty-seven percent of departments say they use agents in some form, but only 17% let them act without a person approving each step. Among departments that let AI act at least at sign-off, 32% allow it to take irreversible financial or legal actions. Nobody owns agent mistakes: a governance or risk team is held accountable in 28% of departments, the deploying employee's manager in 26%, the employee in 25%, the vendor in 9%, and 11% say nobody or do not know.

The numbers

Why it matters

The report's most useful contribution is separating the organisations that get strong returns from the rest, with specific correlates. Departments that judge AI on at least one growth measure (new revenue, conversion, output, customer satisfaction) report strongly positive ROI at 35%, against 13% for those that only count efficiency measures such as hours saved. Organisations standardised on a single approved AI platform report strong ROI at 50%, against 9% for those still experimenting. Departments that monitor their agents automatically have transformed workflows at 24%, against 9% for those that do not.

Two findings cut against popular narratives. First, departments with strongly positive ROI were no more likely than others to report staff reductions or hiring freezes. The AI-pays-for-itself-through-layoffs story does not show up in this data. Second, cost is becoming the constraint: the cost of tools, compute and licences is already the leading barrier to stronger returns, cited by 34%.

Other autumn surveys say similar things. KPMG's Q3 2026 AI Pulse, released on 24 September from 314 leaders at $1 billion-plus companies, found nearly six in ten report measurable business value, 62% are building or deploying agents, and 43% have usage or token budgets. Gartner's CIO survey, released on 19 October, found only 13% of CIOs report significant or transformational value.

What sceptics say

The obvious caveat is that this is self-reported ROI. Ninety-three percent of managers+ say they formally measure ROI at least occasionally, but the report does not audit how. A manager whose budget depends on the AI programme is not a neutral source.

The correlations are also correlations. Organisations with a Chief AI Officer in a new role report strong ROI at 58%, against 8% with no such role planned, but the "none planned" group has only 63 respondents, which the report itself flags as a small base. Well-run organisations probably appoint AI leaders, standardise platforms and measure growth outcomes, and would get better returns from anything.

Finally, the leadership view is rosier than the floor view. Only 58% of entry- and mid-level employees say their department's workflows changed significantly, against 73% of managers+. The people doing the work see less transformation than the people reporting on it.

What it means for engineering and DevOps teams

The engineering-specific findings should make platform teams uncomfortable. Product/Engineering departments use agents about as much as anyone (62%, against 68% for other departments) but only 16% have AI built into their core processes, the lowest of any function, and only 8% met the report's bar for transformed workflows, against 32% of IT departments. Engineers have the tools and have not rebuilt the process around them.

My take, as opinion: that matches what the DORA research on AI and delivery found. Coding agents speed up the individual and shift the bottleneck to review, testing and deployment, which nobody redesigned.

• Pick one measured outcome per AI workflow. Lead time, change failure rate or escaped defects, not "hours saved". The report's growth-versus-efficiency split applies to engineering too.
• Standardise the agent platform. The 50% versus 9% gap is the strongest single correlate in the study. One approved coding agent, one gateway, one set of policies.
• Automate agent monitoring. Human review of agent output did not correlate with transformation; automated monitoring did. Agent observability with OpenTelemetry is the practical starting point.
• Name the owner of agent errors before the first incident, since 11% of departments cannot.

FAQ

What is the Wharton GBK AI adoption report?

It is an annual survey by Wharton Human-AI Research and GBK Collective, now in its fourth year, of employees at large US companies. The 2026 edition, "The Adaptive Advantage: Who's Winning with AI and How", was published on 13 October 2026 and surveyed about 1,400 people in July 2026.

What share of companies get ROI from AI in 2026?

In the Wharton and GBK study, 80% of managers and executives report a positive return and 28% report a strongly positive one. Gartner's October 2026 survey found 13% of CIOs report significant or transformational value, using a stricter definition.

How many companies let AI agents act autonomously?

Wharton and GBK found 67% of departments use AI agents in some form, but only 17% let them act without a person approving each step.

Related: DevOps Still Matters in 2026, Multi-Agent Systems in 2026 and FinOps for AI in 2026.