AI Can Balance Government’s Books. It Can’t Understand What They Mean.

Source Domain: www.governing.com

A county treasurer in rural Texas used to dread the extremely time-consuming manual process of calculating liability for equipment leases under a Governmental Accounting Standards Board (GASB) requirement. After software enhanced by artificial intelligence was adopted last year, the time it took her to do the calculation was cut by two hours — for each lease. Multiply that across dozens of leases and it amounts to real time recovered and real compliance value delivered.

Now consider a different scene. A midsize city uses an AI-assisted forecasting tool to project sales tax revenue. The model produces a confident, internally consistent estimate. The problem: It cannot account for the recent closure of a regional employer, a pending state formula change or the fact that the city’s economic base is more concentrated than its peer group. The number looks authoritative. The judgment behind it is absent.

Both examples involve the same category of technology. Only one of them is doing what the finance officer actually needs.

Most of what has been written about AI in government treats the finance office as just another department. It shouldn’t be. As participants in a recent roundtable in the journal Perspectives on Public Management and Governance argued, public finance must be “recentered” in public administration because fiscal decisions carry legal weight, institutional consequences and accountability demands that generic management tools are not designed to handle. That distinction matters more than ever as AI tools move from IT departments into budget offices.

Three structural differences set the finance office apart. The first is the data problem. Government financial data is fragmented, nonstandardized, backward-looking and structured around fund accounting — not the clean, normalized databases where AI performs well. Unlike private-sector finance, which generates high-frequency…

Source