Board AI Decision-Making Survey Finds CFO Trust Outrunning Governance
Board has found a striking conflict inside the C-suite: 48% of CFOs follow AI advice when it contradicts their own judgment. The Board AI decision-making survey suggests finance chiefs now rely on large language models more readily than CIOs and COOs do. Yet formal controls have not kept pace with that trust.
The shift is bigger than another increase in workplace AI usage. Large language models, or LLMs, are becoming inputs to strategic choices involving budgets, forecasts, hiring, inventory, and investment. They are also displacing sources that executives have trusted for decades.
The central contest is therefore not AI versus human executives. It is executive adoption versus institutional accountability. Businesses are inviting AI into consequential decisions before many can explain, review, or reliably measure its influence.
Board’s AI Decision-Making Survey Reveals a New C-Suite Habit
LLMs have moved from productivity assistants to influential sources of strategic advice.
Board surveyed 300 U.S. CFOs, CIOs, and COOs during May and June 2026. Every participant worked for an organization generating at least $100 million in annual revenue.
The company released its findings on September 16 through the planning intelligence report. Because Board sells enterprise planning software, its conclusions deserve the same scrutiny applied to any vendor-sponsored research. Even so, the role-level differences reveal an important change in executive behavior.
Sixty-one percent of respondents named LLMs among the outside sources that most influence their strategic decisions. That put ChatGPT, Claude, Gemini, and Copilot ahead of every traditional source listed in the survey.
Industry peers and professional networks received 42%. Technology vendors received 37%, while external consultants received 28%. Gartner was cited by 14%.
The result does not prove that executives consider AI more accurate than experienced advisers. It shows that LLM output has become unusually accessible during the decision process. An executive can request a summary, scenario, challenge, or recommendation without scheduling another meeting.
CFOs reported the highest usage among the three roles. Sixty-nine percent cited LLMs as influential resources, compared with 58% of COOs and 56% of CIOs.
The difference becomes more consequential when AI disagrees with the executive. Forty-eight percent of CFOs said they follow the AI recommendation in that situation. The corresponding figures were 33% for CIOs and 11% for COOs.
Across all three groups, 31% said they follow AI advice when it conflicts with their judgment. Only 39% reported formal governance and escalation procedures for AI-driven decisions.
Those numbers describe two related developments. AI has gained a seat in the decision process, while accountability systems still resemble those built for conventional software.
Board CFO Gordon Pothier told CFO Dive that finance chiefs are usually more conservative and skeptical. He argued that their adoption indicates the strength of the current AI push. His more important advice concerned disagreement: executives should pressure-test the recommendation instead of accepting it automatically.
That distinction matters. Consulting an LLM is not the same as delegating a decision. However, following its recommendation during a conflict moves the technology closer to decision authority, even if a human signs the final approval.
Why CFOs Are Turning to LLMs Now
Executives are adopting AI because their planning cycles cannot match the speed of the decisions they face.
Board’s broader results describe an organization struggling with stale information. Eighty-five percent of respondents reported greater pressure to make faster decisions during the previous year. Only 27% said their organizations could re-plan in real time.
Three-quarters said roughly half or more of their planning decisions relied on data over 30 days old. Eighty-three percent said their boards had acted on forecasts known to be outdated. Forty percent reported significant business consequences from those decisions.
These findings provide the missing context for CFO AI decision making. Finance teams are not simply attracted to a conversational interface. They are trying to shorten the distance between a changing condition and an executive response.
Traditional planning often requires collecting operational data, reconciling definitions, updating a model, and preparing a presentation. By the time the revised forecast reaches directors, the assumptions behind it may already have changed.
An LLM appears to offer a faster route. It can summarize reports, compare scenarios, extract assumptions, and frame questions in seconds. Connected systems can also retrieve internal information and incorporate outside signals.
That speed creates genuine value when the task is exploratory. A finance chief might ask which assumptions contribute most to a margin decline. The model can help identify scenarios requiring deeper analysis.
The danger appears when fluent output is mistaken for current evidence. An LLM does not automatically know whether the underlying sales data is complete or properly governed. It can produce a confident explanation from an outdated forecast just as easily as from a current one.
Board’s report contains an uncomfortable sign of this problem. Fifty-nine percent said their AI investment currently exceeds the value it delivers. Yet 92% of that group still planned to raise spending during the following 12 months.
Another 21% said their companies present AI performance more positively to boards, investors, and peers than reality supports. That result suggests adoption pressure is not driven entirely by verified returns.
Pothier summarized the peer pressure with a question: “Do I really want to be the CFO who is not catching onto the wave?” The concern is understandable, but it is not a business case.
Board Chief Product Officer David Marmer offered a more disciplined framing. Buying additional AI, he said, will not solve the underlying planning problem. Organizations must connect AI to specific decisions, assumptions, and accountable people.
This is where an organized AI knowledge base becomes relevant. The model needs traceable context, not an uncontrolled pile of documents. Decision quality depends on knowing which information was used, when it was updated, and who owns it.
CFO AI Decision Making Puts Finance Under New Pressure
Finance chiefs must now defend both the economic value of AI and the integrity of AI-influenced decisions.
CFOs occupy an unusual position in this transition. They approve technology spending, monitor returns, oversee planning, and often challenge optimistic projections. They are now also becoming frequent users of the technology they must evaluate.
That dual role explains why the Board findings matter beyond finance departments. A CFO recommendation influences capital allocation, staffing, pricing, acquisitions, and investor communication. Errors can move across the business before anyone recognizes their common source.
Finance chiefs also face pressure from boards that expect faster answers. An LLM can produce an immediate response when directors ask for another scenario. A conventional planning cycle might take days.
The convenience changes expectations. Once an executive team sees an AI-generated answer in minutes, it becomes harder to defend a slower process. That remains true even when the slower process contains essential validation.
CIOs face a different pressure. They must secure the systems, control access, choose models, and manage technical reliability. Yet Board found that CIOs were less likely than CFOs to cite LLMs as strategic influences.
COOs reported even less willingness to follow conflicting AI advice. Their work often touches physical constraints, supplier conditions, and frontline operations that a general model cannot directly observe.
These role differences create organizational friction. Finance may accept an AI-generated scenario that operations considers unrealistic. Technology leaders may reject a workflow because the model lacks approved access or audit controls.
Board also found uneven adoption of agentic AI, which refers to systems that can plan and perform multi-step tasks with limited intervention. Sixty-six percent of CIOs and 60% of CFOs reported active use in real business processes. Only 38% of COOs said the same.
Reported returns followed a similar pattern. Seventy-three percent of CIOs and 71% of CFOs claimed clear, measurable returns from agentic AI. The figure fell to 52% among COOs.
Those percentages are self-reported and come from vendor-sponsored research. They do not establish a common return calculation across the companies. Still, the gap suggests that each executive function may be evaluating AI through a different operational lens.
The finance organization therefore needs a shared review mechanism. Every consequential recommendation should identify the source data, model, assumptions, and human owner. Teams should also record what changed after the recommendation entered the process.
That record matters when outcomes disappoint. Without it, leaders cannot separate model failure from weak data, poor implementation, or an executive decision that ignored the system.
A useful human role is not merely approving the final output. It is challenging the framing, checking evidence, and deciding whether the question was suitable for AI assistance. Accountability begins before the answer appears.
Adoption Has Moved Faster Than Proven Decision Quality
High AI usage does not establish that AI has improved the decisions carrying the greatest financial consequences.
Independent research reinforces this distinction. A Gartner survey of 204 finance leaders found that 45% of finance AI investments emphasized productivity. Only 20% focused primarily on decision quality.
The finance AI findings point to a gap between activity and enterprise value. Faster document preparation or reconciliation can create useful efficiencies. However, those gains do not automatically improve pricing, investment, or risk decisions.
Productivity is easier to measure. A team can compare hours spent before and after automation. Decision quality often becomes visible only months later, and external conditions can obscure the result.
This measurement problem complicates the Board AI decision-making survey. Executives can report that LLMs influence them, but influence is not an outcome. The survey does not determine whether following a model produced a better decision.
Evidence from the Federal Reserve Bank of Richmond adds useful context. Its corporate AI evidence drew on 603 responses collected in late 2025, plus 145 supplemental responses.
Companies reported benefits in production efficiency, decision speed, and output. However, researchers found little evidence that AI had materially changed overall headcount or costs. Respondents also expected little near-term effect on aggregate employment.
Those results are compatible with meaningful operational gains. They are also less dramatic than claims that AI has already transformed enterprise economics.
Capgemini surveyed 500 C-suite executives in August and September 2025. Only 17% said they actively used AI across multiple stages of strategic decision-making as a standard practice. Another 42% used it selectively.
The executive decision research found that 47% reported noticeable or greater improvements in decision quality. Fifty-nine percent reported similar gains in decision time and cost.
Yet the same study found that only 1% expected AI to make certain strategic decisions autonomously within three years. Executives appear comfortable with assistance, but not with surrendering formal responsibility.
That boundary is easy to describe and difficult to enforce. A leader can remain the official decision-maker while relying heavily on a model’s framing. The human may review the answer without rebuilding the reasoning behind it.
LLMs can also encourage automation bias, the tendency to favor machine recommendations even when contrary evidence exists. Conversational fluency makes that effect harder to notice because the recommendation resembles reasoned human advice.
The Board survey’s conflict question is therefore its most important measure. A CFO who follows AI despite personal disagreement may have discovered evidence they initially missed. The same behavior may also reflect deference to a persuasive but unsupported answer.
The survey does not reveal which explanation applies. That unresolved distinction should prevent readers from treating the 48% figure as either proof of progress or proof of recklessness.
Governance Is the Weak Link in Executive AI Use
The accountability gap becomes most dangerous when AI influences a decision that cannot be easily reversed.
Only 39% of Board respondents reported formal governance and escalation processes for AI-driven decisions. That means executive reliance is expanding while many organizations lack a defined response to conflicting, unclear, or high-risk recommendations.
Governance does not require routing every prompt through a committee. It requires matching the level of review to the consequence of the decision.
Drafting questions for a meeting carries little direct risk. Recommending an acquisition valuation, credit policy, workforce reduction, or earnings assumption carries much more. Those decisions need traceable data and independent review.
The governance problem is not hypothetical. Deloitte surveyed 200 CFOs at North American companies with at least $1 billion in annual revenue. Ninety-three percent said their companies used AI across key operations.
Ninety-six percent expressed confidence in their governance frameworks. Still, 59% identified balancing deployment speed with risk management as their top governance challenge. Cost transparency, cybersecurity, and protected content were also significant concerns.
The AI governance survey found that 19% of CFOs considered themselves primarily responsible for AI governance. That placed finance chiefs behind CIOs and security leaders, but ahead of CEOs and boards.
Confidence in a framework does not show whether employees consistently follow it. Organizations need evidence that controls operate during real decisions, especially when executives feel time pressure.
Capgemini found that 71% of respondents worried about legal and security exposure. Sixty-nine percent cited difficulty explaining AI-influenced decisions, while 60% identified inadequate enterprise data.
Fifty-two percent worried about their ability to verify output quality or relevance. Thirty percent specifically cited the risk of making an incorrect decision based on AI.
These concerns identify the minimum controls for executive use. Models should access approved information, reveal source material, and preserve decision records. A qualified person must challenge the result before action.
Organizations also need an escalation rule for disagreement. If an executive’s judgment conflicts with AI, the answer should not depend on which side sounds more confident.
A review might compare the model’s sources with the underlying records. It could ask a second team to recreate the analysis without seeing the recommendation. High-stakes cases may require legal, security, or domain review.
Companies should also distinguish public LLMs from secured enterprise systems. Entering confidential forecasts into an unapproved service can create privacy and retention risks. Even approved systems can return incorrect answers from incomplete internal data.
Governance must cover the decision context, not just the model. A technically accurate recommendation can still ignore employee effects, contractual obligations, or changing market conditions.
That is why human oversight must involve real authority. A reviewer who cannot delay or reject a recommendation serves as decoration. Effective oversight gives someone responsibility for asking whether the organization should act.
Three Signals Will Show Whether Executive Trust Is Justified
The next phase of enterprise AI will be judged through decision records, measurable returns, and enforceable governance.
The first signal is whether companies begin measuring decision quality separately from productivity. Faster analysis has value, but it cannot remain the main proxy for better judgment.
Finance teams should track a defined set of AI-influenced decisions. They can record the recommendation, the alternative considered, the human rationale, and the eventual outcome. That creates evidence beyond user surveys and vendor claims.
Improved forecast accuracy would strengthen the case for CFO AI decision making. Repeated overrides, corrections, or unexplained errors would weaken it. The relevant measure is not how frequently executives use AI, but how reliably it improves chosen outcomes.
The second signal is whether formal governance catches up with usage. Board found only 39% of executives had established governance and escalation processes. Future surveys should show whether that number rises alongside adoption.
A credible system will define which decisions require human validation and which data sources the model may use. It will preserve a record when AI advice conflicts with executive judgment.
Governance failures would undermine the adoption narrative. These might include confidential data exposure, unsupported financial claims, regulatory scrutiny, or decisions nobody can explain after the fact.
The third signal is whether rising investment produces returns that boards can verify. Board found that 59% believed current AI investment exceeded delivered value. Nearly all of that group still expected higher spending.
That pattern can continue during an early investment cycle. It becomes harder to defend when projects remain disconnected from revenue, risk reduction, forecast accuracy, or measurable operating gains.
Boards should ask whether spending increases because a validated use case is expanding. They should distinguish that reason from investment driven by competitive anxiety.
The Board AI decision-making survey captures a real shift, but not a settled victory for AI. LLMs have become leading sources of executive input before organizations have agreed how to govern their influence.
For knowledge workers, the immediate lesson is equally practical. AI can widen the evidence available for a decision, but its answer should remain connected to reviewable source material. Tools such as ask remio can support that process by grounding questions in a user’s own knowledge.
The test is simple: can your organization reconstruct why an AI-influenced decision was made after the result arrives? If not, the next investment should strengthen context, measurement, and accountability before adding more automated authority.
