Chief AI Officer statistics circulate in a single line: 11% of organisations in 2023, 26% in 2025, 76% in 2026. The line is quoted in board papers, vendor decks and job adverts as evidence that the role has arrived.
Follow each number to its footnote and the line comes apart. Three figures come from three different sources with three different methods, and one of them is not a survey at all. None of that makes the role less important. The tracing changes what the numbers can be used to argue.
What follows traces every widely quoted figure back to where it was produced, says what each can support, and then turns to the one document that defines what a Chief AI Officer actually decides.
76 per cent: a CEO survey of 2,000 leaders, February to April 2026
CAIO adoption at 76% comes from IBM’s 2026 CEO study. The IBM Institute for Business Value ran it with Oxford Economics, surveying “2,000 CEOs and equivalent senior leaders across 33 geographies and 21 industries from February to April 2026”, and reports that “76% of surveyed organizations have a Chief AI Officer (CAIO) in 2026, up from just 26% in 2025”.
Read the respondent. A chief executive answering a survey reports whether the organisation has someone holding the title, not what that person may decide. The release adds that among organisations with the role, “all surveyed CEOs expect the influence of the role to increase by 2030”, which is a statement about expectation rather than current authority.
CAIO surveys would be far more useful if they asked what the role may decide. A question about whether the officer can stop a deployment, grant an exception or control the AI budget would turn a title count into a measure of governance. Interpretation: until surveys ask it, the adoption rate measures fashion in org charts as much as it measures accountability.
What the figure supports: a clear majority of organisations whose chief executives answered this survey report having the title. What it cannot support: any claim about budget control, decision rights or reporting lines, because the release gives none.
26 per cent: a different survey, screening more than 2,300 organisations for a CAIO study
CAIO prevalence at 26% comes from a separate IBM publication, “Solving the AI ROI puzzle”, produced in July 2025 with the Dubai Future Foundation and Oxford Economics. Its method: the team surveyed more than 600 holders of the title “across 22 geographies and 21 industries in the first quarter of 2025”, and in finding them “reached out to more than 2,300 organizations, but only 26% said they currently have a CAIO”.
So the 26% is a screening rate from a study designed to find holders of the title, drawn from 22 geographies. The 76% is a prevalence figure from a chief executive survey across 33. Comparing them tells you both numbers exist, and tells you nothing reliable about the rate of change between two unmatched samples.
CAIO forecasts in the 2025 report add a circularity worth noticing. The same study found that 57% of its respondents “were appointed from the organization’s internal talent pool” and that 66% “expect most organizations will have a CAIO within the next two years”. A 2026 figure showing most organisations with the title is now being read as confirmation of that expectation, although it comes from a different sample again.
Interpretation: “up from 26%” in the 2026 release is a comparison of two different instruments presented as a trend, and any chart plotting the two points on one line inherits the problem.
11 per cent: a 2023 trade article, not a survey at all
Eleven per cent is the CAIO figure most often placed at the start of the line, dated 2023. The 2025 IBM report states the 26% “is up from 11% in 2023”, and attaches footnote 4.
Footnote 4 cites “Weldon, David. ‘Chief AI Officer: What it takes to land the C-suite’s hottest new job.’ CIO. November 9, 2023.” The earliest point on the famous line is therefore a figure reported in a trade publication, carried into an IBM study by citation.
Three points, three sources: a trade article, a screening survey of prospective respondents, and a CEO survey. Interpretation: the only defensible use of the line is to say the title has become common, among the populations these sources describe, since 2023. Any precise growth rate calculated from it is arithmetic on incompatible inputs.
Ten per cent better returns: self-reported, and the method notes say so
CAIO return claims are the second most quoted figures. IBM’s 2025 report states that organisations with the role “see 10% greater ROI on AI spend and are 24% more likely to say they outperform their peers on innovation”, and that those running hub-and-spoke or centralised operating models “see 36% higher AI ROI than those managing decentralized operating models” and move twice as many pilots into production.
Two cautions come from reading the report rather than the headline. First, these are answers from respondents about their own organisations, and organisations that appoint to the role and centralise AI decisions may already be the better-resourced ones, so the figures show the role and the results travelling together rather than one causing the other.
Second, the report’s own notes describe its deeper findings as modelled. The analysis of what drives measurable business impact is “based on multivariate neural network analysis”, and the note adds that the “total predictive power of model with all variables is approximately 40%”. A model that explains roughly two fifths of the variation leaves most of it unexplained.
CAIO return figures are fit for one purpose: justifying a pilot of the role, with the organisation’s own measures set in advance. The figures are not fit to underwrite a business case, because a correlation from other organisations says nothing about what the role will return in yours.
A baseline figure in the same report comes from elsewhere again. “Since 2023, only 25% of AI initiatives have delivered expected ROI” is footnoted to IBM’s 2025 CEO Study, published on 6 May 2025.
Eleven models, sixteen by 2026: a 2024 projection with three months left to run
Scale figures travel with most CAIO articles: “a typical organization uses 11 generative AI models” and “plans to use at least 16 by the end of 2026”. The 2025 report carries it with footnote 5.
Footnote 5 cites “The CEO’s guide to generative AI: AI model optimization”, published by the IBM Institute for Business Value on 29 July 2024. Read in September 2026, the figure is a two-year-old forecast whose horizon is three months away.
Nothing is wrong with the original number. Something is wrong with quoting it now as a current description of enterprise AI estates, and the same care applies to the report’s line that a typical executive “has been offered more than 30 different generative AI use cases from each of their vendors”, which footnote 6 attributes to an IBM study of 10 November 2024.
What IBM’s own model says matters: measurement, teamwork and authority
CAIO success, on IBM‘s analysis, turns on three factors. The 2025 report says its research “points to three key areas where CAIOs who deliver greater measurable business impact with AI focus their attention: measurement, teamwork, and authority”, with the same modelling note attached.
Measurement produced the report’s most uncomfortable figure. “72% of CAIOs in our survey say their organizations risk falling behind without AI impact measurement”, and yet 68% of the same respondents “agree they initiate AI projects even if they can’t measure their effects”. Two thirds of the people accountable for AI start projects they cannot evaluate.
Teamwork produced a structural finding. “The average size of the CAIO team is five, regardless of AI maturity”, and the report states that smaller teams are “less successful”.
Authority is where the report’s language needs reading closely. The report says “CAIOs need a clear mandate to be successful, based on visible support from other C-suite leaders”, ranks the CEO first for support and the CTO first for involvement and advocacy in a structural equation model, and reports that “32% of CAIOs say the CHRO is one of AI’s biggest detractors”.
Authority means sponsorship to IBM and decision rights to OMB
CAIO authority in the IBM report is sponsorship: support, involvement and advocacy from other executives. Nothing in that definition lets the CAIO refuse anything.
Federal authority means something narrower and harder. OMB Memorandum M-25-21, issued on 3 April 2025, lets the officer waive minimum risk practices for a high-impact system “after making a written determination, based upon a system-specific and context-specific risk assessment”, requires annual recertification of each waiver, allows revocation “at any time”, and then states: “The CAIO’s responsibility under this paragraph shall not be delegated down to other officials.”
Same word, two meanings. An organisation can give its officer IBM-style authority, meaning a sponsor at the top table, and still leave the role unable to stop a deployment. Interpretation: when a charter grants “authority”, the useful question is which kind, and the answer should name decisions rather than supporters.
The one source that defines the role: M-25-21, read for what it lets an officer decide
CAIO requirements for federal agencies sit in M-25-21, “Accelerating Federal Use of AI through Innovation, Governance, and Public Trust”, which “rescinds and replaces” M-24-10. Every agency head had 60 days to “retain or designate a Chief AI Officer”, at Senior Executive Service, Scientific and Professional or Senior Leader level at agencies covered by the Chief Financial Officers Act, and at or above Grade 14 elsewhere.
Access is specified. The officer “must be positioned highly enough to engage regularly with other agency leadership, to include the Deputy Secretary or equivalent”. The role may be combined with an existing Chief Information Officer, Chief Data Officer or Chief Technology Officer post, provided that person “has significant expertise in AI”, and a vacancy or change must be notified to OMB within 30 days.
Decision rights are the part to copy. Pilot programmes escape the minimum practices for high-impact AI only where, among other conditions, “the agency CAIO has certified that the pilot may go forward, and that certification is tracked centrally”. Where an official decides that a use case presumed high-impact does not meet the definition, that official “must submit written documentation to notify the CAIO”.
And the power to override runs one way: the officer “may revisit any determinations made within their agency to conclude that an AI use case is considered ‘high-impact'” at any time, which lets the officer escalate a use case into the regime but not release one from it.
CAIO governance boards in the federal model have a prescribed membership, which is worth copying because it forces the right people into the room. Boards must include “appropriate representation from key stakeholder offices or components, including those responsible for addressing IT, cybersecurity, data, budget, statistics, legal counsel, privacy, civil rights, and civil liberties”, and where relevant human capital, procurement, customer experience, programme evaluation and the officials implementing AI in programme offices.
Consequences sit behind the rights. Agencies had 365 days, to 3 April 2026, to document minimum practices for high-impact AI, and where a use case falls short “the agency must safely discontinue use of the AI functionality”. Each agency must also “publicly release a summary describing each individual determination and waiver, as well its justification”, or state publicly that it has none.
Knowledge underpins all of it. Duty E is to “maintain the agency’s AI Use Case Inventory”, and a footnote records that the inventory is kept “As required by Pub. L. No. 117-263, div. G, title LXXII, subtitle B, § 7225 (codified at 40 U.S.C. 11301 note)”. Congress legislated the register, and OMB assigned the officer who keeps it.
Money is the power the federal officer does not hold
CAIO budget powers in M-25-21 are advisory. Duty I is to “provide guidance on AI investments to the agency head and agency CFO related to resourcing requirements”, and duty J is to “support agency efforts to track AI spending”.
Private-sector holders sometimes get more. IBM’s 2025 report found that “61% of CAIOs control their organization’s AI budget”, which is one area where a company routinely gives its officer a power the federal model withholds.
Convening is also structured with care. Each CFO Act agency had 90 days to form an AI Governance Board with “a chair, at the Deputy Secretary level or equivalent, and a vice-chair who is the agency CAIO”, so the officer sets the agenda while someone more senior holds the casting vote.
OMB chairs a separate Chief AI Officer Council drawing members from CFO Act agencies, the Office of Science and Technology Policy and the Office of the Director of National Intelligence, due to sunset five years after the memorandum unless extended.
Scored on four decision powers, to refuse a pilot, to control exceptions, to own the inventory and to hold the budget, the federal model scores three. Interpretation: the federal design deliberately separates the officer who decides on risk from the officials who control money, and a company copying the model should decide consciously whether to follow that separation or to close it.
What Europe asks for instead of a title
CAIO appointments are not required by the EU AI Act, which never names the role. The Regulation requires AI literacy under Article 4, in Chapter I, applying since 2 February 2025.
The Digital Omnibus softened that duty. The original text required providers and deployers to “take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff”. The consolidated text requires them to “take measures to support the development of AI literacy of their staff”, adding: “This obligation does not require providers or deployers to guarantee any specific level of AI literacy of any individual.”
CAIO ownership of literacy evidence is practical rather than symbolic. The records that satisfy Article 4 are training completions mapped to roles, content matched to the systems each team uses, and a view of who operates which AI system, and the last of those is the inventory under another name.
A second requirement uses the word that divides IBM and OMB. Article 26(2) requires deployers of high-risk systems to “assign human oversight to natural persons who have the necessary competence, training and authority, as well as the necessary support”. Article 26 sits in Chapter III, which the Digital Omnibus deferred to 2 December 2027 for Annex III systems, and “authority” there plainly means the power to intervene rather than a sponsor’s goodwill.
Interpretation: European law asks for two owners, one for literacy evidence now and one for oversight authority from December 2027, and a CAIO holding decision rights is the natural candidate for both.
Three numbers worth tracking instead of the adoption rate
A board learns almost nothing about its own CAIO from prevalence figures. Three internal figures tell it a great deal, and each comes from a record the role should already hold.
Inventory coverage: the share of AI systems in use that appear in the register with an owner and a review date inside the last twelve months. Exception discipline: the number of live exceptions, and the share past their recertification date. Measurement: the share of active AI projects with a defined impact measure, which answers IBM’s finding that 68% of respondents start projects they cannot measure.
Each is cheap to produce once the records exist and impossible to produce without them. Interpretation: a board that asks for these three every quarter will learn more about whether its CAIO holds real authority than any survey of other organisations can tell it.
A charter, an inventory, a register: ninety days of evidence
CAIO onboarding plans usually list meetings. A better plan lists records, because every power discussed here is evidenced by a document that exists by day 90 or does not.
By day 30: a charter naming the decisions the officer holds, reserving exception authority personally with no delegation, stating the reporting line and the governance board’s composition, and approved by the chief executive or the board.
By day 60: an inventory covering applications, models, datasets and vendor components, each with an owner and a last-reviewed date, and a documented risk classification method that someone else could apply. By day 90: a pilot certification log, an exceptions register, an AI spend view reconciled with finance, an Article 4 literacy record, and an impact measure recorded against every active project.
CAIO authority is established fastest by using it early. Certifying one pilot, declining or conditioning another, and recording both in the first ninety days teaches the organisation that the role decides things. An appointee whose first quarter contains no recorded refusal will find the first one far harder in month six, because by then the organisation will have concluded the title is advisory.
An exceptions register earns its place by holding five fields derived from the federal waiver paragraph: the system, the requirement set aside, the risk assessment that justifies it, the date recertification falls due, and the person with power to revoke it. The federal text requires annual recertification rather than an expiry date, so the due date is the practical translation.
Governance software earns its place at exactly that point, and turning a CAIO’s decisions into inspectable records is the problem Govern365.ai was built around. Holding the inventory, the high-impact determinations, the pilot certifications, the exceptions register, the literacy evidence and the impact measures in one dated place gives a board the three numbers above without a spreadsheet exercise.
Govern365.ai maps each obligation to the record that proves it across the EU AI Act, ISO/IEC 42001 and the NIST AI Risk Management Framework, so the officer answering all three works from one set of records rather than three that drift apart.
Frequently Asked Questions
What percentage of companies have a Chief AI Officer?
IBM’s 2026 CEO study reports 76% of 2,000 surveyed organisations. The comparison figure of 26% for 2025 comes from a separate IBM study, and the 11% often quoted for 2023 comes from a CIO.com article, so the three are not one series.
Is the growth from 26% to 76% real?
The title has plainly become more common, but the two figures come from different instruments, different respondent groups and different numbers of geographies, so the size of the change cannot be read reliably from them.
Do organisations with a CAIO get better AI returns?
IBM reports 10% greater ROI on AI spend where the role exists and 36% higher ROI under centralised models. Both are self-reported associations, and IBM’s own notes describe its deeper modelling as explaining about 40% of the variation.
What does IBM mean by CAIO authority?
Visible support from other C-suite leaders, with the CEO ranked first for support and the CTO first for involvement and advocacy. Sponsorship of that kind is not the power to refuse a deployment.
Are federal agencies required to have a Chief AI Officer?
Yes. OMB Memorandum M-25-21 of 3 April 2025 required every agency head to retain or designate one within 60 days, at Senior Executive Service level or equivalent for CFO Act agencies and at least Grade 14 elsewhere.
What can a federal Chief AI Officer decide?
Whether a pilot may proceed, whether to waive a minimum practice for a high-impact system, and whether to escalate a use case into the high-impact regime. The waiver power may not be delegated.
Does the federal Chief AI Officer control the budget?
No. The memorandum makes the role advisory on investment and supportive on spend tracking. IBM found 61% of private-sector respondents do control the AI budget.
Can the Chief Information Officer also be the Chief AI Officer?
In federal agencies, yes, provided the person has significant expertise in AI. In a company it is permitted and creates a conflict, because the same person would certify pilots and grant exceptions on systems their own teams deliver.
Does the EU AI Act require a Chief AI Officer?
No. It requires AI literacy under Article 4, now framed as supporting its development, and from December 2027 requires deployers of high-risk systems to assign human oversight to people with the necessary authority.
How big is a typical CAIO team?
IBM’s 2025 report puts the average at five, regardless of AI maturity, and states that smaller teams are less successful.
What should a board measure instead of CAIO adoption?
Inventory coverage with current reviews, live exceptions past their recertification date, and the share of active AI projects with a defined impact measure.
What should a new Chief AI Officer deliver in the first 90 days?
A charter naming the decisions the role holds, an inventory with owners and review dates, a classification method, a pilot certification log, an exceptions register, a reconciled spend view, an Article 4 literacy record and an impact measure for every active project.
