AI in Finance: What’s Real and What’s Hype

AI is one of the most talked-about subjects in finance, and the conversation mixes real capability with a great deal of exaggeration. Claims range from the useful to the fantastical, and separating the two is harder than it should be. Getting it right matters. A finance professional who dismisses AI misses tools that already save time; one who believes the hype risks wasted budget, disappointment, or errors in the numbers.

This guide gives a grounded view for Financial Controllers, Finance Managers, accountants and the businesses that hire them. It sets out why the hype exists, what the published survey evidence says, which uses work reliably today, which claims are not yet real, how to think about where things are heading, and what all of this means for hiring and skills. Where we cite figures, they come from named surveys you can check; where good evidence does not exist, we say so rather than fill the gap.

Why the Hype Exists and Why It Is a Problem

The hype has a real foundation. Large language models have improved quickly and some of what they do is genuinely striking. That real progress is then amplified by vendor marketing, by investors and commentators extrapolating from impressive demonstrations, and by headlines that favour dramatic predictions. A demo that drafts a board paper in seconds becomes a story about finance teams disappearing.

The problem is that hype distorts decisions. Businesses over-invest in tools that do not fit their data or processes, expect savings that do not arrive, or adopt AI carelessly without controls. Equally, the backlash leads some teams to ignore AI altogether and miss modest but real gains. Both mistakes come from the same place: judging AI by claims rather than by what it does on your own work.

What the Evidence Actually Shows

Reliable data on AI in finance functions is thinner than the volume of commentary suggests, and much of it comes from vendor-sponsored surveys. Two independent data points are worth knowing.

• Adoption rose sharply, then levelled off. Gartner’s 2024 survey of 121 finance leaders found 58% of finance functions using AI, up 21 percentage points on 2023. Its 2025 survey of 183 finance leaders found 59%, essentially flat, with Gartner pointing to complexity, data and talent challenges.
• Impact so far is modest. In the 2025 survey, 91% of respondents reported low or moderate impact from their initial AI use. More mature adopters were more likely to report higher impact.
• The common uses are unglamorous. The most-cited uses in 2025 were knowledge management (49%), accounts payable automation (37%) and error and anomaly detection (34%). Finance leaders rated code generation highest for impact, although fewer used it.
• The barriers are skills and data. Gartner found skills gaps and data quality or availability were the largest obstacles.

These are global surveys with modest samples, so treat the percentages as direction rather than precision. The pattern is still telling: AI is common, the useful applications are practical and specific, and most organisations are not yet seeing transformational results. On the audit side, the Financial Reporting Council published its first guidance on AI in audit in June 2025, a sign that the technology is now normal enough to regulate — and that regulators expect it to be documented and controlled.

What AI Can Genuinely Do for Finance Today

Setting the hype aside, five areas of finance work are seeing reliable, everyday value. What they share is that a person can check the output quickly, and the AI handles a first draft or a first pass rather than the final answer.

Drafting

AI is good at turning structured information into readable text. Given a variance table and the reasons behind the numbers, it produces a solid first draft of management commentary; given bullet points, it drafts a board paper section, a policy note or a reply to an auditor. The finance professional still supplies the facts and the judgement, and edits the result. Our guides to AI for variance commentary, AI for management reporting narrative and AI for board pack drafting go into the detail.

Summarising

Reading is a large hidden cost in finance. AI can summarise a long contract, lease, loan agreement or piece of technical guidance, pull out key terms such as covenants, break clauses and payment dates, and answer questions about the document. This is the “knowledge management” use at the top of the Gartner list. The caveat is important: summaries can omit or misstate a clause, so anything that drives an accounting entry or a decision must be checked against the source.

Reconciliation matching

Rules-based matching has existed in accounting software for years. Newer tools add learned matching that suggests likely pairs where references are messy, amounts are split or descriptions differ, and they explain unmatched items in plain language. Good tools speed up bank, intercompany and supplier statement reconciliations; they do not remove the need for a person to approve matches and investigate exceptions. See AI for reconciliations.

Coding and Excel help

This may be the most underrated use. AI writes and explains Excel formulas, builds Power Query steps, drafts VBA macros and short Python scripts, and helps debug a broken model. For an accountant who is strong on finance but not a programmer, that removes a real bottleneck — and it is consistent with Gartner’s finding that finance leaders rated code generation highest for impact. The output still needs testing on known data before it goes near a live model. Our guide to using AI inside Excel covers this in practice.

Document extraction

Extracting data from invoices, receipts, remittances and statements is more accurate than it used to be, because modern tools combine optical character recognition with language models that understand layout and context. It is why accounts payable automation features so prominently in the surveys. Exceptions, new suppliers and poor-quality scans still need human handling, and supplier bank detail changes must never be accepted on extraction alone. See AI for AP and AR.

What Is Exaggerated or Not Yet Real

The autonomous close

The idea of a month-end close that runs itself — AI posts the accruals, reconciles everything, reviews the results and produces the pack — is not current reality. Parts of the close can be accelerated: drafting journals for review, suggesting matches, flagging anomalies, drafting commentary. But accruals, provisions, cut-off and judgemental estimates need someone who understands the business and is accountable for the numbers. Our guide to AI in the month-end close sets out what can realistically be sped up.

Replacing accountants

Claims that AI will imminently replace accountants confuse tasks with roles. AI takes over parts of tasks — the first draft, the first-pass match, the data entry. Roles also involve judgement, accountability, relationships with auditors, banks and boards, and understanding what the numbers mean for the business. None of that transfers to software, and nobody signs accounts on an AI’s behalf. Our article can AI replace a Financial Controller? looks at this for the controller role specifically.

Plug-and-play on messy data

Many promises assume clean, consistent, well-structured data. Most finance functions do not have that: inconsistent coding, duplicate suppliers, spreadsheets outside the system. AI applied to poor data produces confident output built on poor foundations. It is no coincidence that Gartner found data quality to be one of the largest barriers. Fixing the data is often the real project; see data quality and the finance function.

Reliable without checking

The most important limitation is that AI can produce fluent, plausible output that is wrong, and it does not reliably signal when it is unsure. It does not understand a business the way a person does. That makes verification non-negotiable for anything that matters, which is why our guide to human-in-the-loop AI controls treats human review as the core control rather than an optional extra.

The Grey Zone: Agents and Forecasting

Some areas sit between real and hype. AI agents that can take multi-step actions — gathering data from several systems, preparing a reconciliation, drafting a journal for approval — are being piloted, and some will become useful. They raise new control questions because they act rather than advise, and they need approval gates before anything is posted or paid.

Forecasting is similar. Machine-learning forecasts can help with high-volume, pattern-driven lines such as sales by product or cash receipts, and AI can speed up scenario building. But forecasts depend on assumptions about pricing, hiring, customers and the economy, and those remain management judgements. Our guides to AI in FP&A forecasting and AI for cash flow forecasting cover where it helps.

Impact on Hiring and Skills

This is where the real change is showing up for the businesses we recruit for. It is less about headcount disappearing and more about what a good hire looks like.

What is changing in roles

• Less processing, more review. As drafting, matching and extraction speed up, more of the job becomes checking, investigating exceptions and explaining the numbers.
• Analysis and communication matter more. If commentary drafts itself, the value lies in knowing whether the explanation is true and what the business should do about it.
• Systems and data skills rise in value. Power Query, data modelling, ERP knowledge and an understanding of data quality make AI tools work; see our guide for the management accountant facing automation.
• The junior pipeline is a genuine question. Routine tasks have long been how trainees learn. If AI absorbs them, employers need to think deliberately about how juniors build judgement. There is no settled answer yet.

What employers now ask about

In interviews, useful questions are practical: which tools the candidate has used and for what, how they checked the output, an example of an AI error they caught, and where they would not use AI. Candidates who can answer specifically — and who understand data security and their employer’s policy — stand out from those who either overclaim or have not engaged at all. Our articles on how AI is changing what employers look for in an FC and AI finance recruitment cover this from the hiring side.

What has not changed

The fundamentals — technical accounting, controls, business understanding, integrity and the ability to explain numbers to non-finance people — remain the core of every finance role. AI fluency is becoming an expected skill alongside Excel, not a replacement for qualification and experience. Professional bodies are responding with guidance and training, such as the ICAEW’s generative AI guide and ACCA’s technology insights. For director-level hires where AI strategy is part of the brief, sister firm FD Capital handles FD and CFO searches.

How to Think About the Trajectory

AI will keep improving, so today’s limits should not be treated as permanent. Tools that are unreliable now may become dependable, and some grey-zone uses will move into the “works today” column. A sensible finance professional keeps up with new capabilities and re-tests tools periodically rather than forming a view once and keeping it.

The pace and end-point are genuinely uncertain, though. Confident predictions of imminent transformation are as much hype as current exaggerations, and the levelling-off in Gartner’s adoption figures between 2024 and 2025 is a reminder that progress inside organisations is slower than progress in the technology. Expect steady improvement, plan for it, and remain sceptical of anyone who claims to know exactly where it ends.

How to Engage With AI Sensibly

A grounded view leads to a simple approach: use AI for what it does well, keep people responsible for what matters, and verify before relying on anything. In practice:

• Start with a policy. Decide which tools are approved, what data can go into them and what review is required; our AI usage policy guide gives a framework.
• Protect the data. Do not paste confidential or personal data into consumer tools; see AI data security in finance.
• Choose tools on fit. Our comparison of Claude, Copilot and ChatGPT for finance and the AI tool landscape for finance help here.
• Build prompting skill. Better instructions produce output that is easier to check; start with the finance prompt library and prompt engineering for finance.
• Pilot on one process. Measure time saved and errors caught on real work before scaling.
• Be ready for the auditors. Know which processes use AI and how they are reviewed; see what auditors ask about AI.

The wider picture is covered in our AI in finance hub and in where AI helps and where it is dangerous in finance.

Cutting Through the Noise: A Practical Test

When a vendor, a commentator or a colleague makes a claim about AI in finance, a few questions separate substance from noise:

• Is it grounded in what AI does well? Drafting, summarising, matching, extraction and coding help are credible. Autonomy, replacement and transformation claims need much stronger evidence.
• Does it acknowledge the limits? A claim that never mentions errors, verification or data quality is probably inflated.
• Where does the human sit? If the answer is “nowhere”, ask who is accountable when it goes wrong.
• Whose data is the evidence from? A demo on clean sample data says little about your ledger. Ask for a trial on your own records.
• Who paid for the statistic? Vendor-sponsored surveys are not worthless, but independent sources deserve more weight.
• What would the auditors make of it? If you could not explain the control to your audit team, it is not ready for anything material.

A claim that passes these tests is probably grounded. One that fails several is probably hype, however impressive the demonstration.

Frequently Asked Questions

How many finance teams actually use AI?

Gartner’s surveys found 58% of finance functions using AI in 2024 and 59% in 2025, based on samples of 121 and 183 finance leaders respectively. Those are global figures with modest samples, so treat them as indicative.

Will AI replace accountants?

Not on current evidence. AI is taking over parts of tasks, especially first drafts and routine processing, which changes what roles involve. Judgement, accountability and business understanding remain with people, and qualified accountants are still needed to sign off and explain the numbers.

What is the best first use of AI for a finance team?

Low-risk, easily checked tasks: drafting commentary from numbers you already understand, summarising documents, or getting help with Excel formulas. These give quick value and let the team learn how AI makes mistakes before using it on anything material.

Can AI run our month-end close?

It can speed parts of it up — journal drafts for review, reconciliation suggestions, anomaly flags and commentary drafts — but an autonomous close is not a realistic current claim. Estimates, cut-off and sign-off need an accountable person.

What AI skills should finance candidates have?

Practical experience with at least one mainstream tool, the habit of verifying output against source, awareness of data security, and ideally stronger Excel and data skills such as Power Query. The ability to give a specific example of catching an AI error is a good sign.

Why are results often disappointing?

Usually because of data quality, unclear use cases or missing skills — the same barriers Gartner identified. Tools that perform well in demonstrations often struggle on inconsistent real-world data.