AI is not standard software. Excel does not challenge your assumptions. A database does not push back on your framing. AI does. You brief it, it responds, you refine the instruction, it counters, and the work gets better.
Processing Content
In practice, it behaves like an interactive coworker. But unlike any employee in city hall, this coworker is entirely unaccountable. It cannot sign a memo, defend a forecast or answer to a city council.
I made that case on June 30 during a panel at GFOA’s national conference in Chicago titled “Steering Your Artificial Intelligence Strategy through a World of Unlimited Potential and Unreliable Hype.”
On July 7, Abhishek Lodha made a strikingly similar case in The Bond Buyer, describing AI as a “junior analyst that never sleeps.” He was exactly right: the question is no longer whether AI works, but where the human checkpoints go.
As Lodha describes, the buy side already sees the opportunity. AI can ingest thousands of continuing disclosures, extract financial data, compare filings, flag changes and draft first-pass credit analysis.
Issuers face the opposite pipeline.
We do not merely consume market data. We generate the primary source material.
Municipal finance teams produce the budgets, resolutions, ordinances, statutory notices, council memoranda and public presentations that eventually become the disclosures investors consume.
For a lean issuer, the problem is not simply information overload. It is the crushing volume of creating accurate, compliant and understandable public finance documents on deadline.
I became Finance Director for the City of Pontiac, Michigan, on January 20. I inherited a finance office with multiple vacancies and core functions still outsourced, no formalized budget process, and a brand-new city council that needed clear financial information quickly.
This was not a technology experiment. It was a capacity crisis.
Our Deputy Finance Director, Porche Prater, built the quantitative budget framework from the ground up. The AI did not build the budget book. It analyzed historical revenue and expenditure trends and generated draft five-year forecast outputs within that framework.
Our team tested those outputs, refined the assumptions and reconciled the results to source information before deciding what would support the figures ultimately presented in the book. The AI did not make that decision. We did.
This iterative review gave us a more rigorous look at our assumptions, allowing us to present a more concrete, defensible budget.
The AI coworker also drafted initial baselines for budget resolutions, local ordinances and council memoranda explaining variance reports. It drafted public presentations for our millage rate hearings and translated dense requirements under Michigan’s General Property Tax Act into language a new council member or resident could actually follow. It eliminated blank-page syndrome.
As I argued in Chicago, compressing workflows that once took weeks down to a matter of hours gives a lean office an entirely different operational clock. Instead of spending scarce time producing a first draft, we can spend it testing logic, checking statutory requirements, challenging assumptions, improving public explanations and preparing for questions from the council dais.
That is real capacity. It is also real risk.
Lodha warns that an agent’s mistake is not a bad sentence, but a bad decision carried into the next step. Municipal finance officers should take that warning seriously because AI errors rarely look careless. They arrive clean, confident and convincing.
We have the near miss to prove it.
At one point, the AI produced a beautifully formatted draft income tax estimate. The presentation was polished. The reasoning sounded plausible. The number was completely wrong.
Had that estimate moved unsupervised into our five-year general fund forecast or the Mayor’s proposed budget, it would have artificially inflated our revenue projections. By the time actual collections fell short, the city could have been left with an unexpected revenue gap after making spending commitments based on money that never existed.
We caught it because our operating rule is ironclad: the machine drafts, but it never gets to be unsupervised.
Every material number must be traced to its source. Every statutory claim must be checked against the governing authority. Every document must pass through a human checkpoint before it reaches the council, the public or the market.
The hours saved do not disappear. They move from drafting to auditing. For a lean finance team, that is exactly where our time should go.
The question, then, is not whether AI can produce a budget memo, variance narrative, forecast analysis or public hearing presentation. It can. The question is whether the process clearly identifies where the machine stops and human accountability begins.
That line cannot be outsourced.
The AI can work overnight. It can absorb administrative volume. But it cannot sit at the council table. It cannot answer for an error. It cannot sign the final document.
We can give the machine the first draft.
The human must always own the final word.