AI Studies and ROI: Eighty Percent Claim Returns. Seven Percent Clear the Bar #AIg

On October 13, 2026, Wharton Human-AI Research and GBK Collective published The Adaptive Advantage, the fourth annual wave of their enterprise AI adoption study. The headline that travels is generous: 80 percent of enterprise leaders report positive returns from AI. The number that should travel farther is 7 percent, the share whose organizations clear all three of the report’s success markers. GBK’s key findings writeup puts that gap in plain view, and it is a measurement story rather than a model story.

Principal Basil C. Puglisi reads soft ROI claims the same way he reads soft governance claims. His framework HAIA-RECCLIN pairs Reasoning and Dispatch so a Fact, a Tactic, and a KPI sit in front of a named human before the decision ship. Self-reported “positive ROI” is a soft bar. Wharton and GBK defined proving value, transforming workflows, and aligning the organization before they measured who cleared them, and only 7 percent did. That discipline matches RECCLIN more than any slide that grades itself.

What the 2026 wave actually measured

The study surveyed U.S. enterprises with 1,000 or more employees and $50 million or more in revenue, fielded July 9 to 29, 2026. It reached 1,007 managers and above plus 409 entry- and mid-level employees, the first wave in the series to hear from both sides of the house. Wharton directed the research agenda and published the report. GBK designed the instrument and sample, ran the analytics, and generated the insights.

Proving value means strongly positive ROI from a department that formally measures it. Workflow transformation means AI fully embedded in core processes, agents in some form, and significantly changed workflows. Sixty-seven percent of departments use agents and 73 percent say workflows have changed, yet only 20 percent meet all three transformation conditions. Organizational alignment needs four of five factors: adequate AI training, a clear long-term vision, meaningful choice over how and when people use AI, not being held back by people and process rather than the technology, and no real resistance among colleagues. Only 7 percent of organizations clear all three markers.

Where returns strengthen, and where the supervised middle weakens them

Growth measures separate polite returns from strong ones. Thirty-five percent of leaders tracking at least one growth measure report strongly positive ROI, against 13 percent tracking efficiency alone. Ownership matters in the same direction. Fifty-eight percent of organizations with a newly designated CAIO report strongly positive ROI versus 8 percent with no such role, and 50 percent on a single approved platform report strongly positive ROI versus 9 percent still experimenting.

Agents tell a sharper story. Brian Smith of GBK Collective notes on the Wharton page that two-thirds of departments use agents, and only 17 percent let an agent complete a task without a person approving each step. Most companies sit in a supervised middle, paying for autonomy and then reviewing every action, and that is where returns are weakest. GBK’s writeup adds the detail: in 48 percent of departments, AI does the work and a person approves every step. Workflow transformation is weakest there at 16 percent for every-action approval and 18 percent for sample review, and higher at real-time monitoring with guardrails (31 percent) and advise-only (25 percent).

Departments that meet all three transformation components report strongly positive ROI at 67 percent, against 19 percent that do not. Strongly positive ROI departments also changed management structure because of AI at 64 percent versus 47 percent elsewhere, without reporting more staff reductions or hiring freezes than others. Daily use rose 16 points to 62 percent while the share of managers who call themselves excited fell 8 points. Usage and sentiment moved in opposite directions for the first time in the series.

Why Puglisi ties the three markers to HAIA-RECCLIN

HAIA-RECCLIN is not a slogan for “use AI carefully.” Reasoning turns a claim into a Fact that can be checked, a Tactic someone can run, and a KPI someone can watch. Dispatch puts that package in front of a named human who accepts, modifies, or rejects before the next step ships. Applied to this report, proving value is the Fact and KPI pair, workflow transformation is the Tactic that has to be real rather than bolted on, and organizational alignment is whether the people who live with the system can still say yes without being overruled by slideware.

The supervised-agent middle fails that test in practice. Approving every step while buying autonomy is not the same as a checkpoint with binding authority, and it is not the same as automated monitoring with clear guardrails. One path pretends to govern while burning cycle time. The other two paths either keep AI in an advisory role or give agents room under rules a named owner can still defend. Puglisi’s advice is to choose which path you are on in writing, with a name and a review date, before the next ROI deck declares victory.

The training gap makes the same point from the workforce side. Seventy-eight percent of leaders say employees get adequate AI training, against 56 percent of employees. A leader-only dashboard would have missed that gap. RECCLIN Dispatch would not, because the people who run the Tactic are part of the evidence, not a footnote under the scorecard.

What to set before the next ROI slide

Teams that want numbers that survive a skeptic should define the three markers before they grade the program. Write what counts as proving value in the department that measures it. Write what “fully embedded,” “agents in some form,” and “significantly changed workflows” mean in that unit’s language. Write which four of five alignment factors you will hold, and who is accountable when training claims diverge from what employees report.

Then measure growth outcomes alongside hours saved, settle ownership (CAIO or equivalent) and platform policy, and move agent design off the supervised middle unless every-step approval is the deliberate product, not a fear reflex. Sixty-seven percent strongly positive ROI among departments that meet all three transformation components is the report’s clearest signal that transformation, not adoption theater, is what pays.

Eighty percent can claim returns. Seven percent clear the bar the study set in advance. Set those markers before the next ROI slide deck, not after it.

Sources

• Wharton Human-AI Research & GBK Collective. (2026, October 13). The Adaptive Advantage: Who’s winning with AI and how [Special report]. Knowledge at Wharton. https://knowledge.wharton.upenn.edu/special-report/2026-ai-adoption-report/
• Lutz, V. (2026, October 13). Measuring what matters in enterprise AI: Lessons from four years of the Wharton-GBK AI adoption report. GBK Collective. https://www.gbkcollective.com/thoughtleadership/enterprise-ai-study-2026
• Puglisi, B. C. (n.d.). What is HAIA-RECCLIN? Reasoning and Dispatch. basilpuglisi.com. https://basilpuglisi.com/haia-recclin/

Questions readers ask

What is The Adaptive Advantage 2026 AI Adoption Report?

It is the fourth annual Wharton Human-AI Research and GBK Collective study of U.S. enterprise AI adoption, published October 13, 2026. The survey covered 1,007 managers and above plus 409 entry- and mid-level employees at firms with 1,000-plus employees and $50 million-plus revenue, fielded July 9 to 29, 2026.

Why does 80 percent positive ROI still leave only 7 percent clearing the bar?

Self-reported positive ROI is a soft measure. The report defines three success markers in advance: proving value, transforming workflows, and aligning the organization. Only 7 percent of organizations meet all three. A looser definition would look better and teach less about what separates polite returns from strong ones.

Where do AI agent returns weaken in the study?

In the supervised middle, where AI does the work and a person approves every step. That arrangement covers 48 percent of departments. Workflow transformation is weakest there (16 percent every-action approval, 18 percent sample review) and higher under real-time monitoring with guardrails or advise-only designs.

How does Basil C. Puglisi’s HAIA-RECCLIN apply to these findings?

HAIA-RECCLIN pairs Reasoning and Dispatch so a Fact, a Tactic, and a KPI reach a named human before the decision ships. The report’s three markers map onto that discipline: define proving value, make workflow change real, and check whether alignment claims match what employees report before declaring ROI success.

What should leaders set before the next AI ROI deck?

Define the three markers in writing for your unit, track at least one growth outcome beside efficiency, settle dedicated ownership and platform policy, and choose agent autonomy deliberately rather than defaulting to every-step approval. Departments that meet all three transformation components report strongly positive ROI at 67 percent versus 19 percent that do not.

#AIg