The Adaptive Advantage study from Wharton Human-AI Research and GBK Collective, presented on October 13, 2026, asks how businesses measure AI outcomes. GBK describes 1,416 participants: 1,007 leaders and 409 entry- or mid-level employees. The authors report that 80% of leaders see positive returns, while only 7% meet all three defined markers of success together. These are self-reports, rather than an independent financial audit.

One number cannot describe the whole change

In our analysis, a positive assessment can mean different things. Time saved on one task, less repeated work and new revenue are separate outcomes. Combining them into a general impression is unhelpful if an organization wants to understand where its investment works. The first practical step is to define the comparison and its period.

For example, a team might use AI to prepare an initial customer response. We would track time to the final checked answer, required corrections and customer satisfaction. This is our explanatory scenario, not a measured project from the survey. Producing more text is not sufficient evidence of a better service.

Employees provide a separate perspective

The official release reports a training-assessment gap: 78% of leaders consider training adequate, compared with 56% of employees. It also reports strongly positive returns among 28% of leaders. These figures come from the same study and are not a second independent confirmation.

Our interpretation is that training should mean more than access to an interface. We would check whether a person knows how to recognize an error, which data they may use and when to request assistance. If employees spend time repairing unclear AI output, that work should appear in the assessment. Otherwise, the management report may look better than the actual working day.

Controls have a purpose and a cost

NIST's AI RMF is a voluntary approach to managing AI risks during design, use and evaluation. It is not a certificate for a specific product and does not assess the financial returns in this survey. We use it as separate context for management responsibility.

In our analysis, review should match the consequences of an action. Drafting an internal note and changing a customer's order do not require identical procedures. A reasonable workflow identifies who authorizes the action, how its result is recorded and how a mistake is corrected. That also makes the cost of controls measurable, rather than leaving it hidden behind a promise of autonomy.

What we can and cannot conclude

A relationship between a management approach and a better assessment does not establish that the former caused the latter. Teams may differ in their tasks, resources and preparedness. We therefore do not turn the survey into a universal recipe for Bulgarian businesses or promise returns from a particular organizational change.

The useful contribution is a more precise question: what improved, who measured it and what remains outside the report? Our conclusion is that AI adoption and successful use should be tracked separately. The next step for an organization is its own comparison, with clear criteria, preserved limitations and participation by the people who perform the daily work.

Sources: GBK — measuring enterprise AI, 2026 study, GBK — fourth Wharton AI study release, NIST — AI Risk Management Framework.

Editorial verification: 2026-10-14T16:34:39.030582+00:00 (UTC).