AI
AI and the CFO: What Changes in the Finance Function
Artificial intelligence will not replace the finance function. It will change what finance is paid for. Work that can be described will be automated; work that requires judgement will become more valuable.
28 October 2026 · 4 min read
Every few years a technology arrives that is supposed to transform finance. Spreadsheets did; enterprise software did. Usually the transformation is slower and stranger than promised. Artificial intelligence differs in one respect. It can do work that until recently required a trained person to read, write and reason.
Large language models are already useful across a wide range of finance tasks. They draft commentary on monthly results from structured data, summarise contracts, loan agreements and board papers, reconcile transactions and flag anomalies, answer routine questions about policy and produce first versions of models, memos and presentations. None of this replaces a finance professional. It removes a large share of the time spent preparing, so that more can be spent interpreting.
If a task can be described precisely, it will increasingly be done by a machine.
What grows in value
What remains, and appreciates, is the work that resists precise description. Judgement comes first: choosing which of several plausible readings of the numbers is right, and what to do about it. Accountability follows, because someone must sign the accounts, stand behind the forecast and answer to the board, the auditors and the regulator, and a model cannot. Negotiation with lenders, suppliers and investors rests on trust and relationships. Context, the history of a business and the reasons behind past decisions, lives in people. So does the ability to explain numbers to those who are not accountants in a way that changes what they do.
A machine can prepare ten financing structures in a minute. Choosing the right one, and persuading a bank and a client to sign it, remains a deeply human task.
Confident errors
The gravest risk is not that AI will take jobs. It is that it will produce answers that look right and are wrong. These systems can generate plausible figures, references and explanations that bear no relation to reality. In finance, where a small error in an input can become a large error in a valuation or a covenant test, that is dangerous, and the danger grows precisely because the output is fluent and assured.
The answer is not to avoid the tools but to surround them with the controls applied to any important process. Every output needs a named owner who checks and approves it. Figures must be traceable to the ledger, the contract or the source document. Confidential data must stay secure and in the right jurisdiction. There should be an audit trail of what the machine produced, what was changed and who signed it off. Sensible teams begin with drafting and summarising before moving to forecasting and valuation, and they train their people to understand how the tools fail as well as what they can do.
The economics of the change
AI also changes the cost structure of finance. In Intelligence on Tap I argued that intelligence is becoming a utility billed by consumption. For a finance function that means a new line in the budget, the cost of tokens and AI services, set against the time saved. The return will be real but uneven. Teams that redesign their processes around the technology will gain the most; those that bolt it on to old processes will gain little and may add risk.
A sensible adoption takes twelve to eighteen months. It begins by mapping the team's recurring work and the time each task consumes, picking out what is repetitive, rule-based and low-risk. Small pilots follow, on tasks such as monthly commentary or contract summaries, with clear measures of time saved and errors caught. Controls come before scale, with internal audit involved early. Only then should the process itself be redesigned: a monthly close rebuilt around automated reconciliation and drafting can be shorter and more reliable than the old one with software added. The time saved is best reinvested in analysis, business partnering and communication.
Careers and boards
For those starting out, the traditional apprenticeship of reconciliations, data gathering and first drafts will shrink. That is a real problem, because those tasks were how many of us learnt the business. Organisations will need new ways to build judgement in junior staff: more exposure to real decisions, more mentoring, more explanation of why rather than merely what. For experienced professionals the opportunity is to spend more time on the parts of the job that matter. The finance leader of the next decade will manage a team of people and machines and will be judged on the quality of the decisions that team supports.
Boards should be asking where AI is already used in their finance function and who approved it, how the accuracy of AI-assisted reports is verified, what data is being shared with outside providers and on what terms, how junior staff will develop judgement if routine work disappears, and what measurable return the spending has produced so far.
AI will not end the finance function. Preparation becomes cheap; judgement, accountability and trust become the job. The teams that grasp this early will be worth more than ever.
