Deloitte released the latest edition of its State of AI in the Enterprise report on October 29, 2026, and two of its numbers belong side by side. Forty percent of companies now allow AI systems to execute decisions or take actions without human approval, at least to a moderate degree. Only 15% report a significant shift in decision-making authority as a result of AI. The survey covered 3,500 director to C-suite leaders across 25 countries in May and June. The two figures measure different things, and read together they suggest that AI has gained authority to act faster than the people beside it have gained authority to decide.
Basil C. Puglisi reads that gap as a return problem as much as a governance problem. In October 2025 he argued in The Real AI Threat Is Not the Algorithm. It’s That No One Answers for the Decision that “In properly governed systems, humans make decisions. AI provides intelligence that helps humans decide better.” His position for clients follows from that line, and it reaches the return on AI as much as its risk. A company that gives its people better information and leaves their decision rights where they were has paid for speed it cannot easily convert into returns. A company that lets AI act with no named owner has a return that nobody answers for.
Deloitte finds work moving faster while authority stays put
The speed gains are broad, with 81% of companies reporting faster decision-making and execution, according to Deloitte. Only 25% say they are using AI to transform the business at its core through new products, reinvented core processes or changed business models. Sixty percent have redesigned jobs around AI to at least a moderate degree, which is four times the share reporting a significant shift in decision-making authority. The share of companies reporting revenue gains from AI rose to 28%, up from 20% in Deloitte’s January report, and Deloitte says the gap between AI’s faster benefits and its slower financial returns is narrowing.
The release explains why the authority number matters. Employees “may have access to better information and make faster decisions, but they aren’t empowered to act differently,” and Deloitte says effective redesign requires changes to roles, approval processes and decision rights. The full report works through that argument at greater length. Jim Rowan, Deloitte’s US head of AI, put the task plainly in the release: “The next challenge is safely turning speed into meaningful change.” Deloitte sells AI consulting, so a client should weigh its prescriptions alongside that commercial interest and read the survey figures as self-reported by the leaders who answered.
Two other studies released the same week report modest returns
Dun & Bradstreet’s quarterly AI Momentum survey of 10,000 businesses in 32 countries, released October 28, found that 28% report broad or strong returns from AI, unchanged from the prior quarter. That is a different measure from Deloitte’s revenue figure, even though both land on 28%. Sixty-five percent plan to increase AI spending over the next twelve months, and just 9% say most of their pilots have scaled with clear gains. The company provides commercial data for enterprise AI, and its release ties returns to data readiness, so that framing carries a commercial interest.
The UK government’s AI adoption survey, published October 29 from 3,503 telephone interviews conducted between January and April, found that half of adopting businesses reported a productivity increase. Those gains were generally described as modest, and on operating costs 64% had seen no immediate financial effect. The same survey found that 81% of businesses apply some level of human oversight to AI outputs, a share that had declined slightly since 2025. Its summary concluded that responsibility for outcomes remained firmly with users. These three studies ask different questions of different samples, and their percentages should not be compared directly. Deloitte’s revenue figure moved up while Dun & Bradstreet’s return figure held still, so the surveys do not agree on direction. In each of them, clear financial gains come from a minority of respondents.
An AI action without an owner leaves its return without one too
Deloitte’s release adds an ownership finding to the autonomy figure. Even as 40% of companies let AI execute decisions without human approval to at least a moderate degree, 23% say accountability remains unclear when an AI system produces a negative outcome. Deloitte’s own advice is that governance should match the level of autonomy and consequence, with more oversight for systems that execute actions than for systems that only recommend.
Puglisi’s 2025 essay described that failure in plain terms: “When an algorithm produces an error, there is no one to answer for it.” The same gap affects the upside. A gain from an AI-executed action needs a named owner who can say whether the action was right and whether it should continue. Without that owner, the gain is hard to defend in a budget review. Automated monitors and spending alerts are machine checking machine, which Puglisi classifies as Responsible AI, and they belong in every deployment. Deciding whether an AI-executed action stands is AI Governance, and under his Checkpoint-Based Governance that decision belongs to a named person who records accept, modify or reject.
How Puglisi would redraw decision rights in an AI-enabled workflow
The first step is a list of the decisions in each AI-enabled workflow, separate from the list of tasks. For each decision, the team records who makes it today and whether AI only recommends or actually executes. It then records who should make that decision now that AI supplies the information, which is often a person closer to the work than the current approver.
The second step gives every AI-executed decision type a named outcome owner. That person sets the conditions under which AI may act alone, and any action outside those conditions returns to the owner for a recorded accept, modify or reject. A company in Deloitte’s 23% with unclear accountability can close that gap one decision type at a time.
The third step tests each legacy approval step against the judgment it adds. An approval where the approver weighs evidence and records a ruling stays in place. An approval that only adds a signature to a decision someone else already made moves to the person who made it. A human still decides, and the decision stops waiting on a second signature.
A client can track the change with two numbers. The first is decision latency, the time from an AI output to the human decision that acts on it, measured for each workflow before and after the redesign. The second is the share of AI-executed decision types with a named outcome owner, and the target is all of them. Neither number is a return by itself. Each workflow’s existing financial measure, such as cost per case or revenue per account, belongs beside them over the same period, so the client can see whether faster decisions are producing value.
Budget season will test whether faster work turns into returns
Sixty-five percent of businesses in the Dun & Bradstreet survey plan to increase AI spending over the next twelve months. If Deloitte’s pattern holds, much of that money will buy speed inside job designs whose decision rights haven’t changed. A redesign that changes the tools and leaves authority where it was keeps the old pace of decisions. Puglisi’s position is that the return from AI lands with whoever is allowed to act on what it produces. The next editions of these surveys will show whether the 15% share reporting a significant shift in decision-making authority moves, and whether the share with unclear accountability falls as more companies let AI act.
Sources
- Deloitte. (2026, October 29). Enterprise AI enters an ROI reckoning: More use, uneven value, reveals Deloitte survey [Press release]. https://www.deloitte.com/us/en/about/press-room/enterprise-ai-enters-an-roi-reckoning.html
- Deloitte AI Institute. (2026). The state of AI in the enterprise [Report landing page]. https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/content/state-of-ai-in-the-enterprise.html
- Deloitte AI Institute. (2026). The state of AI in the enterprise: Built differently [Report]. https://www.deloitte.com/content/dam/assets-zone3/us/en/docs/services/consulting/2026/the-state-of-ai-in-the-enterprise-2026.pdf
- Department for Business, Innovation, Science and Trade. (2026, October 29). AI adoption survey report 2026: Executive summary. GOV.UK. https://www.gov.uk/government/publications/ai-adoption-survey-2026/ai-adoption-survey-report-2026-executive-summary
- Dun & Bradstreet. (2026, October 28). Dun & Bradstreet survey of 10,000 businesses finds AI returns stable for most, but data-ready companies pull ahead on ROI [Press release]. PR Newswire. https://www.prnewswire.com/news-releases/dun–bradstreet-survey-of-10-000-businesses-finds-ai-returns-stable-for-most-but-data-ready-companies-pull-ahead-on-roi-302922158.html
- Puglisi, B. C. (2025, October 18). The real AI threat is not the algorithm. It’s that no one answers for the decision. basilpuglisi.com. https://basilpuglisi.com/the-real-ai-threat-is-not-the-algorithm-its-that-no-one-answers-for-the-decision/
Questions readers ask
What did Deloitte’s State of AI in the Enterprise report find in October 2026?
Deloitte’s survey of 3,500 leaders found that 81% of companies report faster decision-making and execution, while only 25% are using AI to transform the business at its core. The share reporting revenue gains from AI rose to 28%, and 40% of companies allow AI to execute decisions without human approval to at least a moderate degree.
Why are AI returns lagging behind AI speed gains?
Deloitte’s release points to unchanged decision rights as one constraint. Sixty percent of companies have redesigned jobs around AI, but only 15% report a significant shift in decision-making authority. Basil C. Puglisi’s view is that faster information produces little value when the people receiving it can’t act on it.
Who is accountable when an AI system makes a bad decision?
Many companies have an answer, but Deloitte’s October 2026 survey found that 23% say accountability remains unclear when an AI system produces a negative outcome. Under Basil C. Puglisi’s Checkpoint-Based Governance, each AI-executed decision type has a named owner who decides whether an action stands and records accept, modify or reject.
Are other surveys finding the same pattern in AI returns?
Dun & Bradstreet’s October 2026 survey of 10,000 businesses found 28% report broad or strong AI returns, unchanged from the prior quarter. The UK government’s 2026 survey found half of adopters reported a productivity increase, generally modest, and 64% had seen no immediate effect on operating costs.
How can a company measure whether decision rights are holding back AI returns?
Track decision latency, the time from an AI output to the human decision that acts on it, for each workflow before and after redesign. Also track the share of AI-executed decision types that have a named outcome owner, with a target of all of them, and report both beside the workflow’s existing financial measure.
#AIg
Leave a Reply