Public finance professionals have long safeguarded public money through budgeting, accounting, financial reporting, internal controls and audit. But a harder question remains unanswered: is public spending actually achieving what it was meant to achieve? In other words, are public resources delivering public value in services, infrastructure, security, social protection, economic growth and improved wellbeing? As governments face pressure to show evidence for their choices, the gap between financial stewardship and policy effectiveness is becoming impossible to ignore.
The International Public Sector Accounting Standards Board (IPSASB) is turning its attention to climate-related outcomes of public policies following the publication in January, of its first climate-related disclosures standard relating to public sector entities’ own operations.
This second phase is where public policy and public financial management (PFM) meet, and where AI could ultimately turn a long-standing ambition into common practice.
Efficiency first
AI is increasingly being used in financial and accounting processes. Optical character recognition lifts data from invoices, while machine learning supports reconciliations, transaction classification, cash forecasting and anomaly detection. These are tools for pattern-spotting and automation.
Audit is moving too. Intosai, the global body for supreme audit institutions, has pointed out that AI lets auditors test entire transaction populations, rather than samples, and turn unwieldy material such as contracts and reports into usable evidence.
AI can help public money do what it is meant to do
Useful as this capability is, it mostly sharpens efficiency and control. The larger prize lies elsewhere. Most reforms of the past two decades have improved the control and reporting of public money, but the harder and more valuable step is understanding whether that money actually changes outcomes.
Hidden evidence
Governments hold enormous quantities of evidence: budgets, financial reports, evaluations, audit findings, performance data and administrative records. Yet this data is scattered across institutions and rarely informs day-to-day decisions. The newer tools can bring it into play.
Generative AI and large language models can read this unstructured material, summarise it and let professionals question it in plain language. In addition, machine learning can find patterns in structured data, including which interventions are linked to which results. Together, these tools have the potential to link financial and performance information.
Both generative AI and machine learning are tools already in use. In the UK, the government’s Consult AI tool has used large language models to theme thousands of free-text consultation responses at a scale manual review could not match (see also the government’s Humphrey AI tool).
On the analytical side, one study used machine learning to sift more than 1,500 climate policy measures across 41 countries and identify those with the strongest links to lower emissions.
One tool makes sense of text, the other finds ‘what works’ in the data.
Opportunity, not threat
Closing the gap between financial management and policy creates opportunities for professional accountants in the public sector. Much of the profession’s effort has gone into collecting, checking, reconciling and reporting. As AI absorbs more of that routine load, accountants’ real strengths come to the fore: judgment, analysis and advice.
ACCA’s research on the changing world of work makes the same point for the profession, describing the accountant’s role as moving from guardian of knowledge to interpreter and trusted adviser, with competence in AI joining analytical and critical thinking as the skills that ACCA members rate highest for the future. The public sector needs that same blend.
AI is a powerful assistant, not a decision-maker
None of this reduces the need for professional judgment. If anything, it raises it. The risks differ by tool: generative models can be fluent yet confidently wrong, while analytical models can embed bias. The barriers are practical too, since connecting fragmented public records is as much a data quality and governance problem as an AI one, and weak data produces confident but misleading conclusions.
AI is a powerful assistant, not a decision-maker. The moment we let an opaque model set spending priorities without human accountability, we have lost the very thing public finance exists to protect. Intosai takes the same view, favouring a hybrid model that pairs machine analysis with human judgment and accountability.
The real question
The opportunity is not simply to automate what we already do. It is to help public finance answer one of government’s most important questions: are public policies, and the money behind them, making a real difference? Technology will not fix weak institutions or replace political will. But, if we use it carefully, AI can help public money do what it is meant to do, and work better for everyone.
That is what the IPSASB’s deliberations on public policy outcomes is now focused on. Policy and public finance are no longer separate conversations. Used with care, AI can help those who manage public money move from confirming how it was spent to understanding what it achieved.
