AI and the CFO: Standing Behind the Numbers the Machine Produces

EY’s 2026 Global DNA of the CFO Survey asked more than 1,600 finance leaders how ready their function is to use AI well. Only 21% rated their preparedness as leading or advanced. Over the same period the function was treating AI as central to its future: in Deloitte’s Q4 2025 CFO Signals, 87% of CFOs expected AI to be extremely or very important to finance operations in 2026, and 54% named integrating AI agents a top priority for the year.

The gap between those numbers is the working condition of the role. Expectation and commitment run well ahead of readiness. CFOs are clear that AI will matter to finance, and a majority are already making its integration a stated priority for the year, yet only a fifth judge their function equipped to use it well. The desire to adopt has outrun the work of building the conditions adoption requires.

A capability gap is usually a roadmap problem, where given enough time and investment, it closes. Accountability is not like that. The CFO is answerable for the integrity of the accounts and the stewardship of capital every reporting cycle, ready or not, and AI is already inside the work that produces both. The doing of that work can move to a machine. The accountability for it cannot. What matters in the meantime is where, duty by duty, AI is already changing how the job is done, which is what this article works through: where it does the work, where it sharpens the judgement, where it operates out of sight, and where it changes almost nothing.

The principle the CFO operates

The modern finance remit is broad, but two duties define the office. The first is the integrity of the financial record: the accounts give a true and fair view, and the CFO stands behind them. The second is the stewardship of capital: the CFO answers for whether the organisation’s spending earns its keep. Almost everything else the function does serves one or the other.

Both rest on binding texts. The Companies Act 2006 requires every company to keep adequate accounting records and to prepare accounts that give a true and fair view. The FRC’s 2024 UK Corporate Governance Code requires the directors to present a fair, balanced and understandable account of the company’s position, to assess its principal risks, and to maintain and review its internal control. The recognised shape of the role was captured in Deloitte’s Four Faces of the CFO framework, which describes the steward, operator, strategist, and catalyst, and corroborated by the ICAEW, which sets out a remit spanning leadership, operations, controls, and strategy. None of these has been rewritten for AI.

Within that settled remit, agency and accountability behave differently. A model can perform much of the work involved in the close. A tool can draft the forecast. An agent can flag a control breach. None of those uses is illegitimate in itself. What cannot move is accountability for whether the numbers are true and the capital was well stewarded. The texts have not changed. The conditions under which the duties are discharged have changed completely.

Where AI does the work, and a human still signs

The first group is the routine numbers work, including the close, reconciliations, tax preparation, and the drafting of standard reports. This is where AI does the heavy lifting well.

The sharper point comes first, though. Finance understands human-in-the-loop governance better than almost any function in the organisation. Segregation of duties, approval thresholds, attestation, reconciliation, and independent review are disciplines the function has built over decades. They are precisely the controls that AI oversight requires. The sensible pattern in practice gives a tool autonomy over gathering, analysis, and drafting, while a person approves anything that touches the ledger or leaves the building. Finance has been operating that pattern long before AI arrived. The control muscle already exists.

This is where the readiness gap needs reading with care. EY’s finding that only 21% feel ready measures AI-specific capability and tooling, not control discipline. The discipline is there. It has not yet been pointed at AI. That is a more encouraging position than the headline number alone suggests, and it is the opportunity in this group: the function that already knows how to govern a human-in-the-loop process is well placed to govern a machine in one.

What AI is not doing is signing. No CFO is putting a true and fair view over a close that an agent ran unsupervised. Deloitte’s State of AI in the Enterprise (2026), drawn from 3,235 leaders across 24 countries, found that worker access to AI rose by 50% in a year, while only 25% of organisations had moved 40% or more of their AI pilots into production. Adoption is broad; production is early and uneven. Autonomous execution of the close is an intent, not a practice.

The line holds. The sign-off on the accounts stays human. The Companies Act fixes the true and fair view as a matter the directors certify, and the FRC Code requires reporting that is fair, balanced and understandable. A machine can produce the figures faster and with fewer errors — it cannot be the one who certifies them.

Where AI sharpens the judgement, and the call stays human

The second group is where AI upgrades the work rather than replaces the worker. Forecasting, scenario analysis, capital allocation, and risk assessment all depend on reading an uncertain future, and this is where the technology earns its place. A finance team that once hand-built three or four cases can now map a far wider range of possible futures and test each against live data. Combined with the four indicator types set out in earlier work, which separate the confirmed, the signalled, the modelled, and the proven, this moves the forecast closer to maximum fidelity: everything knowable made available before the judgement is made.

One discipline matters more than the tooling. Predictive work explores a portfolio of possible scenarios; it does not name a single likely outcome. The value sits in the range, not in the point estimate. A Board that reads a distribution of futures and seizes on one path as the answer has misused the instrument. The range is the finding. The pull to collapse it into a forecast is the error to guard against.

Better forecasting serves the Board directly. It sharpens the capital-allocation decisions the CFO and the Board own, and it supports the thorough assessment of principal risks the FRC Code requires of the directors in Provision 28. The same modelling gives the audit committee a clearer view of where the genuine uncertainties lie, which is where independent challenge does its real work.

The call does not transfer. The instrument improves; the judgement does not move with it. A wider, better-modelled set of futures makes the allocation decision better informed. It does not make the decision for anyone. The choice, and the accountability for it, stay with the CFO and the Board.

Where AI operates out of sight, and the CFO still owns the consequence

The hardest group is the one the CFO can see least. Two things are happening at once. AI is entering the control environment itself, the very system of checks the financial record depends on. And AI is being bought and used across the business, creating value and risk in places the finance function has no clear line to.

The common reading is that the CFO is the executive least consulted on AI. That is the wrong frame. The CFO is the officer accountable for the financial consequence of AI that the organisation is deploying faster than anyone, the CFO included, can see or govern. Material financial consequences are increasingly being created by AI systems embedded throughout the organisation rather than by systems owned directly by the finance function.

Where deployment decisions are made deliberately, ownership is often contested rather than clearly the CFO’s. KPMG’s 2025 CFO and CIO Collaboration Survey, a small sample of around 100 US leaders at billion-dollar companies, found 59% of CFOs and 61% of CIOs each claiming primary responsibility for AI investment decisions. The figure is illustrative rather than definitive, but it reveals a familiar problem: where ownership is shared, accountability can become blurred.

Visibility is the second challenge. A 2025 Gartner survey found 69% of organisations suspect or have evidence of staff using prohibited public AI tools. Much of the value and risk is now being created outside formal governance processes. When that hidden activity results in a security incident, a compliance failure, or a poor investment decision, the financial consequence still lands on the CFO’s desk. IBM and the Ponemon Institute found shadow AI was a factor in 20% of breaches and added roughly $670,000 to the average breach cost.

The opportunity is not to slow adoption. It is to extend control and assurance to AI itself, and to ensure the value it creates is measured with the same discipline as any other investment. Minimum Lovable Governance is the operating principle: light enough to keep the speed advantage, structured enough that the audit committee can attest to what has been built. The same control discipline that governs a human-in-the-loop process extends to the machine, including the growing practice of finance teams generating code that touches audited systems, where the verification premium decides whether that output can be trusted.

The line to hold is the sharpest in the piece. Accountability follows the consequence, not the line of sight. The task is to build the visibility, because the CFO is answerable for the value and the risk whether or not the function can currently see them. The FRC Code makes this concrete in Provision 29, which requires the directors to monitor and review the effectiveness of all material controls, with the effectiveness declaration applying for financial years beginning on or after 1 January 2026. A control environment the CFO cannot see is a declaration the CFO cannot truthfully make.

The part that does not move

Some of the role barely changes at all. Three parts of it sit almost untouched by AI. The first is the judgement of whether the business is a going concern, a matter Cadbury placed at the constitutional core of the finance function in 1992. The second is the relationship with the audit committee and the external auditor, the structure of independent challenge set out in the FRC Code. The third is the act of standing behind the numbers to investors and to the market, which the directors’ responsibility statement makes a personal undertaking. These are the moments when the CFO’s two duties become tangible. The going-concern judgement asks whether the business can continue, and the audit relationship tests whether the numbers can be trusted. Standing behind those numbers is where accountability becomes personal.

These are matters of judgement and relationship. AI can inform them at the edges. It can surface an early signal of liquidity stress, or assemble the evidence an audit committee reviews. It cannot form the judgement that the business will meet its obligations as they fall due, and it cannot sit inside the relationship of trust between a CFO, an audit committee, and an auditor. Naming this plainly matters, because the anxiety around AI tends to assume the whole role is dissolving. The core of it is not. The parts of the job that turn on judgement, and on standing behind a position, are the parts AI reaches last, if it reaches them at all.

The line that does not move

This is the same principle the series has carried from the start, applied now to the finance chief. AI and the Director, AI and the Chair, and AI and the Company Secretary traced it through the oversight roles, whose duty is to the Board’s collective accountability. The CFO is the first executive officer in the series, and the object shifts to the integrity of the financial record and the stewardship of capital. The principle does not shift with it. Agency for the doing can be transferred to the machine. Accountability for the truth of the numbers and the stewardship of the capital cannot.

The readiness gap is real, and it is the tension the CFO works inside. It is not the ending. The ending is older and steadier than any survey finding. The CFO’s job has never been to produce the numbers. Clerks, then systems, then models have always produced them. The job has been to stand behind them. AI changes who produces the numbers. It does not change who signs for them. That is the line that does not move.