AI In Public Finance

Author: Asante Nxumalo

AI in Public Finance: From Public Money to Public Intelligence

Imagine a weekday morning near the end of 2026. A public-finance analyst arrives at work. Before the first meeting, an AI assistant has already summarised a dense policy document, compared expenditure against budget, surfaced anomalies in a procurement dataset and prepared a first-pass briefing note. Work that once swallowed an afternoon has taken minutes.

The speed is impressive. It also creates a more difficult set of questions. Can the analyst trust the summary? What data was the model trained on? May sensitive government information be entered into it? If the system flags a supplier, forecasts a fiscal pressure or recommends a spending priority, who is accountable for what happens next?

That tension is the real story of artificial intelligence in public finance. The question is no longer whether AI will touch the work of the state. It is whether public institutions can build the capability, governance and imagination to use it in service of better public outcomes.

Public finance is bigger than government accounting

Public finance is the machinery through which a state raises resources, decides how they should be allocated, spends them through public institutions and accounts for what was achieved. It stretches across tax and revenue administration, budgeting, procurement, expenditure management, public investment, debt, public entities, reporting, audit and oversight.

Behind those systems sits a stubborn reality, there is only one fiscus. Health may need more money because patient demand is rising. Education may need classrooms in fast-growing communities. Transport may need to maintain ageing roads. Municipalities may face revenue pressure while infrastructure backlogs deepen. Each claim can be legitimate, but scarce resources still have to be prioritised.

Public finance has therefore never been only about recording what has already happened. At its best, it is about seeing pressure early, testing choices before they are made, protecting public resources and understanding whether spending is translating into outcomes. This is where AI becomes more interesting than a faster spreadsheet.

From automation to institutional intelligence

The most immediate uses of AI are familiar. It can summarise documents, classify information, search large bodies of material, prepare first drafts and reduce repetitive administrative work. Those gains matter, especially in institutions where scarce professional capacity is consumed by locating, cleaning and organising information.

The larger opportunity, however, lies beyond automation. AI can support forecasting, pattern detection, scenario analysis and decision support across the fiscal cycle. It can help institutions move from asking only what happened to asking what is changing, what might happen next and where human attention is most urgently required.

Consider a health department combining expenditure data with patient volumes, medicine procurement, staffing information and historical demand. A forecasting system identifies that several facilities are likely to exhaust part of their operating budgets before year-end. The warning appears in September rather than February. The value is not that an algorithm has made a budget decision. The value is that officials have time to investigate the cause, test options and respond before the pressure becomes a crisis.

The same principle can be applied elsewhere. In education, scenario modelling can help analysts compare the consequences of delaying infrastructure, protecting spending in high-growth districts or shifting lower-priority projects into later years. In infrastructure, financial and operational data can be read together to identify projects where expenditure is rising faster than physical progress. In procurement and audit, anomaly detection can help professionals focus limited investigative capacity on transactions or patterns that warrant closer scrutiny.

Parts of this future have already arrived. The OECD Tax Administration 2025 report notes that 69% of surveyed tax administrations used AI in 2023, with a further 24% in the process of implementing it. Tax administrations are using AI across analytical work, taxpayer services, case selection and repetitive administrative tasks. In a domain defined by enormous volumes of declarations, payments, correspondence and supporting information, AI can make the haystack more searchable. It can help surface patterns and prioritise attention. But a risk score should not quietly become a verdict. Officials still need to understand why a case was selected, interrogate the evidence and apply the law.

This is the conceptual shift that matters. The promise is not merely a government that processes the same work faster. It is a government that can connect information across functions, recognise emerging risks sooner, model choices more intelligently and direct scarce human capacity toward the questions that require judgment.

Public money changes the stakes

Efficiency is only one value in public finance. Legality, fairness, transparency, privacy, auditability, inclusion and democratic accountability matter too. A flawed recommendation in an ordinary workflow may waste time. A flawed recommendation embedded in a public-finance process can affect services, livelihoods, rights and trust.

Data quality is the first constraint. AI cannot rescue fragmented, incomplete or biased information. It can simply process those weaknesses at greater speed and scale. Opaque models create a second problem whereby officials need enough understanding to challenge outputs, explain decisions and establish who is responsible. Automation bias creates a third. A recommendation that appears quantitative can acquire authority it has not earned.

Privacy and security raise the bar further. Public institutions hold sensitive taxpayer, employee, supplier, commercial and citizen information. That makes questions about access, data location, cybersecurity, vendor dependency and the ability to audit external systems central to responsible adoption. For South African institutions, these are not abstract technology questions. They are questions about institutional control and, in some cases, data and AI sovereignty.

A useful governing principle: automate the process where appropriate, augment the professional wherever valuable and preserve human accountability wherever public judgment is exercised.

This principle is consistent with the direction of the IMF's 2026 work on emerging digital technologies in public financial management. The starting point should not be technology for its own sake. Institutions should begin with clearly defined public-finance objectives, assess where digital capabilities can improve outcomes, and then test those opportunities against practical feasibility, governance and institutional readiness.

South Africa's window

South Africa enters this conversation at an important moment, and a cautionary one. In March 2026, Cabinet approved the publication of a draft South Africa AI Policy for public comment. The draft was gazetted on 10 April 2026 and withdrawn on 26 April 2026, after at least six of its 67 academic references were found not to exist. The department attributed the failure to AI-generated sources that were published without verification. Its stated direction included capacity and talent development, inclusive growth and job creation, responsible governance, ethical and inclusive AI, cultural preservation and international integration, and human-centred deployment. These principles matter for public finance because good AI adoption depends as much on how institutions are run as on the technology itself. The withdrawal shows why. The AI policy document written to govern AI in our country was discredited and withdrawn due to using AI without checking the result, and South Africa is now left without a national AI policy.

The potential applications are not difficult to imagine. Revenue authorities can strengthen compliance analysis and taxpayer support. Treasuries can improve forecasting, expenditure monitoring and fiscal-risk analysis. Municipalities can examine billing anomalies, revenue leakage, procurement patterns and infrastructure maintenance needs. Oversight institutions can interrogate larger transaction populations and concentrate professional capacity where risk appears greatest. Public entities can combine financial, operational and governance information to identify deterioration earlier.

But AI is not a substitute for fixing structural governance problems. Weak controls do not become strong controls because a model sits on top of them. Fragmented data does not become reliable because it is processed more quickly. Institutions that lack internal capability can become more, not less, dependent on vendors they cannot adequately interrogate. An AI-ready public-finance institution therefore needs more than software. It needs reliable data, fit-for-purpose infrastructure, skilled people, clear governance, disciplined procurement, secure experimentation and leadership that understands both the opportunity and the limits. The sequence of experiment, validate, govern and then scale, matters.

There is also a public-facing opportunity. Fiscal reporting is often dense even when the underlying information is technically available. Responsible AI-enabled interfaces could allow citizens, journalists, legislators and civil-society organisations to interrogate public financial information in plain language to better observe What changed? Why did it change? What was spent? What was achieved? What does this mean for my community? Better transparency will still depend on better data and institutions, but the distance between public information and public understanding could become shorter.

The intelligent state is not an automated state

Return to the public-finance analyst a few years later. The first meeting of the day still involves trade-offs. There is still uncertainty. Departments still compete for scarce resources, and no model can decide which social priority deserves greater moral or political weight.

What has changed is the quality and timeliness of the information around those decisions. Revenue risks are detected earlier. Budget scenarios can be modelled rapidly. Procurement anomalies are surfaced before losses crystallise. Financial and operational information can be read together to show where spending is failing to translate into outcomes. Officials spend less time searching through reports and more time interrogating what those reports mean. Citizens can ask clearer questions of public financial information.

The final image is therefore not an autonomous machine deciding where a country spends its money. It is a more capable state. Humans remain responsible for judgment, trade-offs and democratic accountability, while AI expands their ability to see, test, anticipate and act.

AI is moving at extraordinary speed. The opportunity for the public sector is not simply to keep up, but to shape and deploy these capabilities in ways that improve how public resources are raised, allocated, protected and translated into public value. The future of public finance should not be artificial intelligence replacing public judgment. It should be better-informed public judgment, strengthened by intelligence that institutions know how to govern.

To continue the conversation, we invite you to engage in the AI Fundamentals AllegroED course, where we unpack the foundations of artificial intelligence, responsible adoption, practical uses of AI and why understanding the technology is becoming increasingly important for professionals and organisations across both the public and private sectors.

Selected sources and further reading

Bochner, R. (2025, November 24). Artificial intelligence as public financial management infrastructure. IMF PFM Blog. https://blog-pfm.imf.org/en/pfmblog/2025/11/artificial-intelligence-as-public-financial-management-infrastructure

Department of Communications and Digital Technologies. (2024). South Africa national artificial intelligence policy framework. https://www.dcdt.gov.za/sa-national-ai-policy-framework/file/338-sa-national-ai-policy-framework

Organisation for Economic Co-operation and Development. (2025). Tax administration 2025: Comparative information on OECD and other advanced and emerging economies. OECD Publishing. https://doi.org/10.1787/cc015ce8-en

Pattanayak, S., Rivero del Paso, L., Tourpe, H., & Cho, C. (2026). Harnessing emerging digital technologies toward a new frontier of public financial management (Technical Notes and Manuals No. 2026/006). International Monetary Fund. https://doi.org/10.5089/9798229040440.005

South African Government. (2026, April 2). Statement on the Cabinet meeting of 25 March 2026 and Special Cabinet meeting of 1 April 2026. https://www.gov.za/news/cabinet-statements/statement-cabinet-meeting-25-march-2026-and-special-cabinet-meeting-1-april

South African Government News Agency. (2026, April 26). Minister announces withdrawal of draft AI Policy. https://www.sanews.gov.za/south-africa/minister-announces-withdrawal-draft-ai-policy