On October 5, KPMG published the third-quarter edition of its global survey on artificial intelligence, with responses from 2,131 executives in 20 countries collected between July 23 and August 26 [1]. Its central thesis is that competitive advantage is shifting from adopting AI to managing it well, which the firm breaks down into four capabilities: governing where models come from, securing what they can reach, coordinating how they are used, and accounting for what they cost.
The figure behind that thesis is striking. Among organizations reporting an established return on investment, 86% operate a formal AI management layer, cross-functional or enterprise-wide, and 65% have deployed it across the whole organization. Among those still experimenting, those proportions fall to 31% and 10% [1].
Reading the correlation with care
The tempting conclusion is that governance produces the return. It may, but the survey does not prove it. A company that is already getting results has reasons and resources to formalize its management, so causality may run the other way or both ways. It is also worth remembering that these are figures declared by the executives themselves and gathered by a firm that advises on the subject. None of this invalidates the finding; it only calibrates it.
The US chapter, published on September 24 with 314 executives from organizations with annual revenues of at least $1 billion, shows what that management consists of [2]. 74% already include a cost review when approving each AI use case, up from 61% the previous quarter. 70% use monitoring dashboards and 43% have set usage or consumption budgets. And 49% have defined high-risk cases in which agents are not allowed to decide autonomously.
The uncomfortable figure
The same chapter contains a combination that deserves more attention than the headline [2]. Executives' confidence in their ability to govern AI jumped from 57% to 73% in a single quarter. Organizations building, developing, or deploying agents rose from 53% to 62%, and those developing or implementing multi-agent systems reached 25%, compared with 6% in each of the two preceding quarters. At the same time, the share of companies building controls into their agents, alongside monitoring and evaluation procedures, fell from 43% two quarters earlier to 30%, because many rely on the safeguards they already had.
More agent deployment, more confidence, and a smaller share of companies reporting controls built into agents. It may be an efficient decision: reusing what already works instead of duplicating it. It may also be the profile of the periods that precede an incident, when experience accumulated without mishap is mistaken for safety. The survey does not reveal which reading is correct, and that is precisely why every management team should know which one applies to its own company.
Three decisions that do not require a department
A mid-sized company does not need a multinational's structure to apply the essentials. It can start with three written decisions. Who approves each use of AI, and with what cost estimate. What is measured every month, with cost and result in the same table. And which decisions an automated system never takes without a person who signs.
The third deserves particular attention: according to the survey itself, just under half of the large US companies consulted have formalized it [2]. There is nothing technophobic about it. It is the same logic as the signing powers and limits every company already applies to its employees, extended to a new kind of collaborator. It is an application of the Judgment Reserve: the trained and organized human capacity an entity maintains to design, supervise, and govern what AI produces and does.
A company that does not know which decisions it would not delegate to a machine cannot be sure which ones it is already delegating.
Sources and methodology note
- KPMG International. Global AI Pulse Q3 2026. AI at scale: Accountability, resilience and economics. October 5, 2026.
- KPMG LLP. AI's Value Story Sharpens as Organizations Gain Confidence in Governance, Accountability and Workforce Adoption. September 24, 2026.
The global figures [1] and those from the US chapter [2] have different sample scopes; the US sample forms part of the global research. They should not be treated as two independent studies or compared directly.
© 2026 Javier F. Pérez Bernabé. All rights reserved. Published by Quórum 12.
ONLINE SUBSCRIPTION
Receive every new analysis and invitation to open events.
This is a separate public category: it does not confer membership, admission or access to the private portal.