The short answer
Use AI to widen the evidence and speed up the analysis, and keep the call with a named person.
In a randomized study of 758 consultants, those using AI finished 12.2 percent more tasks, 25.1 percent faster and with better quality inside its strengths, but were 19 percent less likely to get a task right outside them.
Decide in advance which decisions AI may make alone, which it only informs, and which are never delegated.
In this article 6 sections
The five lessons
Read these if you read nothing else.
- AI improves decisions inside its strengths and makes them worse outside them. You have to know which is which before you rely on it.
- Most CEOs are already comfortable deciding on AI input (64 percent, IBM 2026). Few initiatives return what was expected (25 percent, IBM 2025).
- Write the decision rights down: what AI may do alone, what it only informs, what is never delegated.
- Test AI on real past decisions before live ones, and set the kill rule before the test starts.
- Give the board one page: every AI use in production, its owner, its number and its last result.
What the research says about AI and leadership decisions
Newest first. Predictions are labelled. The experiment ran in 2023 and was published in a peer-reviewed journal in 2026.
| Finding | Sample | Source |
|---|---|---|
| 64 percent of CEOs are comfortable making major strategic decisions based on AI-generated input; 79 percent of executives are decentralising decision making | 2,000 CEOs and senior leaders, 33 geographies, February to April 2026 | IBM Institute for Business Value, May 2026 |
| Inside AI's strengths: 12.2 percent more tasks, 25.1 percent faster, higher quality. Outside them: 19 percent less likely to be correct | 758 BCG consultants, preregistered randomized experiment, run in 2023 | Dell'Acqua et al., Organization Science, 2026 |
| 60 percent of companies report minimal revenue and cost gains from AI | 1,250 senior executives, 2025 | BCG, September 2025 |
| Only 25 percent of AI initiatives delivered expected ROI; 16 percent scaled company wide | 2,000 CEOs, 33 countries, February to April 2025 | IBM Institute for Business Value, May 2025 |
| Prediction: over 40 percent of agentic AI projects will be cancelled by the end of 2027 | Gartner analysts, poll of 3,412 webinar attendees, January 2025 | Gartner, June 2025 |
| CEOs rate only 44 percent of their CIOs as AI-savvy; 66 percent say their business models are not fit for AI | 456 CEOs and senior executives, June to November 2024 | Gartner, May 2025 |
| 31 percent of boards do not have AI on the agenda; 17 percent discuss it at every meeting; 66 percent say the board does not know enough about AI | 695 board members and executives, 56 countries, January to February 2025 | Deloitte Global Boardroom Program, 2025 |
- The experiment is the most useful number here. The same people, with the same tool, did better on some tasks and worse on others, and could not always tell which was which.
- The CEO surveys show comfort running ahead of results. That gap is a governance problem, not a technology one.
The decision-rights table
Fill this in once with your leadership team and review it every quarter. The test for each row is simple: if the decision goes wrong, who answers for it, and how fast can it be undone?
| Who decides | Kind of decision | Examples |
|---|---|---|
| AI may act alone | Small, frequent, easy to undo, with a clear rule | Routing a new lead to the right person; sending the instant reply; flagging a spend or reply-time anomaly |
| AI recommends, a named person decides | Repeated decisions with money attached, where evidence helps | Moving budget between channels; which open quotes to chase first; which test to run next month |
| A person decides, AI only researches | Rare, large or hard to reverse | Pricing for a key account; entering a new market; choosing a vendor for a multi-year contract |
| Never delegated | Decisions about people, the law or the company's word | Hiring and firing; legal and contract commitments; anything said publicly in the company's name |
How to test AI on a real decision before you trust it
Use your own history. It is cheaper than a pilot and shows where AI is inside or outside its strengths for your business.
- Pick one repeated decision with a known outcome, such as which quotes closed last quarter.
- Give AI the same information your team had at the time, and nothing that came later.
- Compare its calls with what your team decided and with what actually happened, case by case.
- Set the bar before you look: for example, it must match or beat the team on at least 20 past cases.
- If it clears the bar, run it side by side on live decisions for four weeks, with a person still deciding.
- Write the kill rule now: if it is no better than the team after the live month, it goes back to research only.
What the board should see
Deloitte found that 31 percent of boards do not discuss AI at all. A one-page register fixes that without a committee.
- Every AI use in production, with the decision row it sits in from the table above.
- The named owner of each one, and the number it is meant to move.
- Last month's result against that number, and any incident or override.
- What it can reach: which systems, which customer data, which spending limit.
What to check this week
List the AI tools your teams already use, then place each one in a row of the decision-rights table. Any tool that acts alone on money, customers or anything public needs a named owner and an approval step this week.
For the operating model behind it, see governed AI operations. For where to start, see the AI implementation roadmap.
Questions leaders ask
How do CEOs use AI for decision making?
Mostly to gather and test evidence faster. In IBM's 2026 study of 2,000 CEOs and senior leaders, 64 percent were comfortable making major strategic decisions based on AI-generated input. The call should still sit with a named person.
Does AI improve decision quality?
Inside its strengths, yes: in a randomized study of 758 consultants, AI users finished 12.2 percent more tasks, 25.1 percent faster, with higher quality. Outside them, they were 19 percent less likely to be correct (Dell'Acqua et al., 2026, experiment run in 2023).
Which decisions should never be delegated to AI?
Decisions about people, legal and contract commitments, and anything said publicly in the company's name. AI can research them, but a named person decides and answers for them.
What is the role of AI in leadership?
To widen the evidence, shorten the analysis and flag what changed. Leadership keeps the decision rights, the accountability and the judgment about when AI is outside its strengths.
How should a board oversee AI?
With a one-page register: each AI use in production, its owner, the number it should move, last month's result and what it can reach. Deloitte's 2025 survey found 31 percent of boards do not have AI on the agenda at all.
SOURCES
- IBM Institute for Business Value, 2026 CEO Study (May 2026): 2,000 CEOs and senior leaders; 64 percent comfortable deciding on AI input
- Dell'Acqua et al., Navigating the Jagged Technological Frontier, Organization Science (2026; experiment 2023): 758 consultants
- BCG, AI Leaders Outpace Laggards (September 2025): 1,250 senior executives
- IBM Institute for Business Value, 2025 CEO Study (May 2025): 2,000 CEOs; 25 percent of initiatives delivered expected ROI
- Gartner (June 2025): prediction that over 40 percent of agentic AI projects will be cancelled by end of 2027
- Gartner (May 2025): 456 CEOs; 44 percent of CIOs rated AI-savvy
- Deloitte Global Boardroom Program, Governance of AI, 2nd edition (2025): 695 board members and executives
Where’s the money leaking?
Pick one. Be honest.
Cite this article
Grigorchuk, T. (2026, October 17). AI for Decision Making: What a CEO Can Hand to AI, and What Stays With You. Megawebvision. https://megawebvision.com/insights/ai-for-decision-making