The real impact of AI in Finance
Finance is often discussed as if it were one job. It is not. The accountant closing the monthly books, the analyst rebuilding a forecast, the auditor testing controls and the investment manager assessing a company carry different responsibilities.
AI is spreading across the field, but its value and risks change with the work. It can prepare, compare, search and classify at speed. The closer a task gets to an important decision, the more finance knowledge and human judgement matter.
The finance function is already using AI, although not evenly
KPMG surveyed finance executives at 2,900 companies across 23 countries and territories. Its 2025 report found that 71% of companies were using AI in finance operations, with 41% using it to a moderate or large degree. Adoption covered accounting, financial planning, treasury, risk and tax.
Look at a specific function and the picture changes. The Association for Financial Professionals reported in its 2025 FP&A research that only 23% of respondents used AI daily, weekly or monthly. A company can be “using AI” while most of its finance team is still experimenting or waiting for approved tools.
In accounting and reporting, AI can classify transactions, reconcile records, identify unusual entries and draft commentary. It reduces the time spent assembling a first version. It does not decide whether the treatment is correct, an exception is material or the final account gives a fair picture. Faster production makes review more important because plausible explanations can appear before anyone has challenged them.
FP&A has a similar split. AI can compare actuals with budget, find movements across thousands of rows, generate scenarios and draft monthly commentary. The analyst has more time to ask why the variance exists and what management should do about it. That only works when the model receives the right definitions and data. Revenue, margin, churn and headcount often mean different things across systems. AI can calculate the wrong metric impeccably if you give it the wrong context.
Treasury gains from faster cash forecasting, payment monitoring and exposure analysis. Tax teams can search legislation, organise documentation and flag anomalies. The OECD’s Tax Administration 2025 found that 69% of surveyed tax administrations were using AI in 2023, with another 24% implementing it. Their examples included analysis, taxpayer services and case selection. Tax conclusions still depend on jurisdiction, timing and facts a general model may not know.
Audit and risk show why speed is only part of the value
Internal audit is adopting generative AI quickly. The Institute of Internal Auditors reported in March 2025 that the proportion of North American chief audit executives using it for audit activities had risen from 15% to 40% in a year.
Applications include document review, risk assessment, test planning and report drafting. The larger opportunity is coverage: teams can inspect more transactions and direct attention towards unusual ones. Yet an audit trail must show how a conclusion was reached. If AI cannot explain a flag, or misses a problem because its data treats it as normal, efficiency has not improved assurance.
The same applies to fraud, credit and compliance. Pattern recognition can surface suspicious payments or exposures. A professional still has to understand false positives, bias and the consequences of acting. A financial error can affect a customer, filing, lending decision or allocation of capital.
Financial services bring the decision closer to the customer
Banks are using AI in customer service, fraud detection, lending, compliance and operational processes. Insurers use it across claims, pricing, underwriting and customer interactions. EIOPA reported in February 2026 that nearly two-thirds of surveyed European insurers were already actively using generative AI, although most reported use cases remained at proof-of-concept stage. The most common applications were internal productivity tools, with human oversight retained in the large majority of cases.
A policy summary creates one level of risk. A model influencing cover, credit or a fraud warning creates another.
Investment management is changing more behind the scenes than in the investment mandate itself. CFA Institute noted in 2025 that only 0.01% of 44,000 EU UCITS funds explicitly referred to AI or machine learning in their formal strategy. Managers were nonetheless using AI to support research, productivity and human-led decisions. It can read filings, compare companies, test signals and summarise markets. It cannot carry fiduciary responsibility or explain away a poor decision by pointing to a model.
There is also a system-wide issue. The Bank of England warned in April 2025 that shared weaknesses in widely used models could cause firms to misprice risk or allocate credit badly at the same time. Similar AI-driven trading behaviour could worsen market stress. At scale, many similar decisions can become a financial stability concern.
The work left to people becomes more valuable
Across finance, AI moves effort away from collecting information and producing a first draft. More time can go into checking data, challenging the result and deciding what happens next.
Those are not soft extras around the “real” technical work. They are what make financial work dependable.
The skills gap is visible. ACCA’s 2025 survey of more than 10,000 accountancy and finance professionals found that half were concerned about developing future skills, while only 32% said their organisation provided opportunities to learn AI-related skills.
Giving a finance team access to an AI assistant will not close that gap. People need to practise with their actual forecasts, reconciliations, controls and risk reviews. They need to see where the output fails and when to accept, improve or reject it.
The impact of AI in finance will be measured in faster work, wider analysis and better decisions. Whether organisations achieve all three depends less on access to a model than on the capability of the people using it.
Etiq helps finance professionals build that capability through personalised learning and practice based on real work. It checks prompts, data grounding, calculations and outputs as people work, giving teams evidence of what they can do well and where they still need support.
