Only 61% of executives surveyed by IBM say AI decision rights are consistently applied
Seventy-three percent of enterprise AI leaders say their organizations have defined the decisions where human oversight is non-negotiable, while 61 percent say decision rights for AI are clearly defined and consistently applied.
A global survey by the IBM Institute for Business Value and Oxford Economics found 75 percent of respondents say governance, risk, or compliance delays AI-driven action.
North America accounted for 19 percent of respondents, Europe 30 percent, APAC 28 percent, Latin America 9 percent, the Middle East and Africa 8 percent, and Japan 6 percent.
Respondents included chief AI officers, chief data and AI officers, heads of AI and equivalent AI leaders.
Eighteen percent of respondents say they have full control over AI investment decisions, and 15 percent have full authority over AI workforce strategy.
Joanne Wright, senior vice president, transformation and operations at IBM, writes in the study’s foreword that “transformation can’t be mandated. It must be earned.”
Forty-eight percent of respondents say AI agents initiate and execute actions across workflows.
Among the subset the study labels high-impact CAIOs, 96 percent say AI-generated outputs can trigger actions autonomously, compared with 40 percent of all other respondents, and the group is nearly four times as likely to say it can quickly pause, adjust or redirect AI systems without disrupting operations.
Kazunari Tokunaga, general manager, DX promotion department/DX at Daikin Industries, Ltd., tells the IBM Institute for Business Value that “it is necessary to design processes that explicitly incorporate human checkpoints.”
A separate IBM study on AI sovereignty published in June, cited within the report, found 57 percent of technology executives have full visibility into their AI stack dependencies, and 71 percent say switching their primary AI vendor would be difficult.
Twenty-nine percent of all respondents say decisions and resources are adjusted quickly when AI performance measurement identifies opportunities to improve outcomes while 40 percent say decisions are discussed but action is delayed.
Thirty-one percent say thresholds for stopping AI initiatives are critical to AI success.
Among high-impact CAIOs, 96 percent say they reduce or stop funding to initiatives that fail to deliver expected outcomes, even where those initiatives have senior sponsorship, against 46 percent of all others.
Ninety-four percent use business outcomes to determine which initiatives receive additional funding and talent, and 95 percent say AI-generated productivity gains are reinvested into higher-value transformation initiatives rather than absorbed into baseline operations, against 44 percent of all others.
Pascal Belaud, executive vice president and chief AI and data officer at Truist, is quoted in the report: “We stop when the value isn’t there.”
The study measured high-impact CAIOs against an index built from three outcomes: the share of AI initiatives generating measurable business results, the share moving from pilot to production, and the time required to scale successful initiatives enterprise-wide.
That group reports 63 percent of projects delivering measurable business outcomes, against 43 percent, and scales a successful initiative in eight months, against 13.5 months.


