If you work in local government finance, you've probably heard that AI is going to change everything.
It will reconcile accounts, find errors, answer questions about your financial data, read invoices, write reports, and save your team hours of work. Depending on the source, it can sound like we're not far from having AI run the finance office.
But we're not there yet, and I don't think that should be the goal anyway.
Before I moved into public sector software, I spent four years as a governmental auditor. One of the questions we were trained to ask was simple: Who reviewed this, and how do you know it's right?
AI doesn't make that question obsolete. If anything, it makes it more important.
That's how I think finance directors should approach AI. Not by asking whether they should be "using AI," but by asking a much more practical question: What work can AI help my team do better today, and where do I still need a person making the decision?
For a small finance team already juggling the general ledger, payroll, utility billing, accounts payable, reporting, and probably a few things that aren't technically in anyone's job description, that's the conversation worth having.
Start with the work, not the technology
One of the mistakes I see organizations making with AI is starting with the technology.
They see an impressive demonstration and then start looking for places to use it. I'd turn that around.
Start by looking at the work your team does every day. Where are people spending time looking for exceptions? Where are they re-keying information? What takes hours to analyze? What repetitive tasks keep getting pushed to the end of the day?
Those are the places I'd look at first.
AI is particularly useful when it can review a large amount of information, recognize patterns, identify exceptions, summarize what it finds, and help someone decide what deserves attention.
There are already several places where that can be useful in a local government finance office.
Accounts payable
Think about how much of the accounts payable (AP) process is still spent reading and entering information from invoices.
AI can read an invoice received by email or scanned into the system, identify the vendor, invoice number, amount, and due date, and prepare that information for review. As these systems get better, they can also suggest general ledger accounts based on previous invoices, identify potential duplicates, recognize unusual amounts, and route invoices through the appropriate approval process.
The time your staff spends typing information from one place into another can go to the transactions that need their attention.
Bank reconciliation
A similar opportunity exists with bank reconciliation, where most transactions match easily, and the time goes into the exceptions.
Software can help suggest matches, identify timing differences, find amounts that are close but not exact, flag stale transactions, and group items that may be related.
Instead of someone spending an afternoon looking through hundreds or thousands of transactions trying to find the problem, AI can help narrow the list to the handful that deserves investigation. Whoever is working on the reconciliation gets to those items sooner and still decides what happened and how to fix it.
Financial analysis
Financial analysis may turn out to be one of the most useful applications of AI for finance directors.
Most ERP systems are very good at storing financial information, but getting answers out of it still takes work, and a question that sounds simple can mean running several reports, exporting them to Excel, and working through the numbers by hand. With AI built into your financial software, you could ask questions like these in plain language:
- Why is Public Works over budget this year?
- Which departments are running materially ahead of last year's spending?
- Are there accounts with unusual activity this month?
- Where is overtime increasing, and what's driving it?
AI can analyze the underlying information and point you toward the areas that deserve attention. You'd still check its explanation before acting on it, but you'd start that review with a much better idea of where to look.
Payroll and utility billing
Outside the general ledger, AI can flag payroll changes that fall outside normal patterns, such as unusual overtime, unexpected changes in deductions, significant pay changes, or an employee setup that doesn't look like similar employees.
In utility billing, it can identify unusual consumption, possible meter-reading problems, unexpected adjustments, or accounts whose usage suddenly looks very different from their history.
These are good uses of AI because the software isn't making the final decision. It's saying, in effect, "You might want to look at this."
For a small team responsible for thousands of transactions and accounts, that kind of early flag can catch a payroll error before checks go out, or a leak before a resident opens a water bill several times higher than usual. Those are the problems that turn into phone calls, corrections, and sometimes questions from council, and catching them early saves everyone the cleanup.
Writing and summarizing
Many finance staff have already started here. AI can prepare a first draft of a council memo, summarize a long policy document, turn financial information into a plain-language explanation for residents, draft answers to common questions, or help write a budget narrative. A draft you edit is often faster than starting with a blank page.
It's one of the easiest places for a municipality to begin, since the output is easy to review and the risk stays low as long as employees know what information they should and shouldn't put into the tool.
The line isn't whether AI touches the work
AI is going to touch more of the work in a finance office, so I don't think the right dividing line is "AI work" versus "human work." It can analyze information, recognize patterns, recommend next steps, draft explanations, prioritize exceptions, and increasingly take actions within software.
The more important question is who owns the decision.
If it involves spending public money, changing financial records, determining compliance, answering to council, or exercising judgment about a resident or employee, a person should remain accountable for the final call.
That matters even more in finance, because "almost right" isn't good enough. A number that is right 95 percent of the time isn't a number a finance director can rely on. The expectation is that the information is accurate, the result can be explained, and there's a record of how you got there.
So I believe the most valuable role for AI in local government finance is to augment the work people do. AI can find the exception, analyze thousands of transactions, recommend an action, or prepare much of the work behind a decision, but we still need appropriate controls and human accountability around the result.
That's also how we're thinking about AI as we build the next generation of products at Caselle. Trust, accuracy, explainability, and auditability have to be at the center. In government finance, you need to be able to trust the answer AI gives you, understand where it came from, and know who reviewed or approved it.
AI doesn't change that standard. Finance departments already rely on software to calculate payroll, post transactions, calculate utility bills, and produce financial statements. We don't manually reproduce every calculation to prove the software is right. We build controls around the system, review exceptions, reconcile results, and maintain accountability, and AI should be approached in much the same way.
Caselle connects your accounting, payroll, and utility billing so every number is accurate, traceable, and audit-ready.
Explore Caselle Software
Good AI needs good data
AI is only as useful as the data it can see, which makes connected systems one of the most important parts of getting value from it.
If the general ledger is in one system and utility billing is somewhere else, and employees are moving information between them with spreadsheets and manual entries, even a very capable AI tool is working with an incomplete picture.
Before investing heavily in AI, I'd look at the foundation. Are your systems connected? Is information entered once and available where it's needed? Are you still re-keying data between applications? Can you trust the information in the system?
Fixing those problems, whether through better integration between systems or a data cleanup and migration project, saves your team time on its own, and it gives any AI tool you add later cleaner, more complete information to work with.
Protect your data and your controls
There's an important difference between AI that's securely built into your business software and a free public AI tool an employee finds online. Bank information, payroll data, Social Security numbers, and resident account information are all data that shouldn't end up in a public chatbot.
That's why every organization should have a policy covering the use of AI. At a minimum, employees should understand what tools they're allowed to use, what information they can provide to those tools, what outputs require review, and what decisions can't be delegated to AI.
And remember that AI creates risks outside your own organization as well. Fraudsters have access to the same technology. More convincing phishing emails, fake invoices, impersonation attempts, and cloned voices make your existing controls around payments and vendor changes more important, and no AI tool should be allowed to shortcut them.
Ask better questions of your software vendors
As AI becomes part of more government software, finance directors are going to hear a lot of promises. Whether a product "has AI" matters much less than how that AI works, and a few questions will tell you a lot:
- When the AI gets something wrong, how will we know?
- Where does our data go, and is it used to train a shared model?
- What information can the AI access, and can access be controlled based on an employee's role?
- What can it recommend, and what can it do on its own?
- Is there an audit trail showing what the AI did and what a person approved?
- What happens when the AI isn't confident?
- Can we turn the feature off?
The answers will tell you much more than an AI checkbox on a feature list.
Where a small finance team should start
If I were sitting down with a three-person finance office tomorrow, I'd skip the AI strategy committee and the big technology project and start with one problem, something repetitive, time-consuming, and easy to verify.
That might be drafting responses to common resident questions or summarizing documents, or it might be a feature already in your financial software that flags reconciliation exceptions, explains budget variances, or handles part of your AP process.
Try it for a few weeks with the usual review step in place, and keep track of the time. If something that used to take two hours now takes thirty minutes, including the time spent checking the result, it's worth keeping. If it isn't saving time, drop it and move on to the next problem.
Approached that way, AI becomes a series of small improvements to individual parts of the job, which is a much easier thing for a small local government to take on than one large transformation project.
Where AI in finance is headed
We're still early. Over time, I expect AI to become less of a separate feature you open and more of a capability built into the software you already use.
Your financial system will point out something unusual in the budget and explain why it deserves a look. When an invoice comes in, the system will read it, suggest how to handle it, and bring in a person when something doesn't look right. And when you have a question about your numbers, you'll be able to ask the system directly rather than clicking through several screens and reports.
That's a much bigger change than adding a chatbot to an ERP system, but the core responsibility of the finance office stays the same. AI already has a place in local government finance, and that place is going to grow quickly. What matters is using it where it makes your team better without weakening the controls and accountability you've spent years building.
In my experience, the municipalities that get the most from new technology, AI included, are the ones with connected data, sound processes, and a clear sense of where it helps their people (and where a person still needs to make the call).
Subscribe to the Caselle newsletter for practical guidance on local government finance, technology, and the tools that make your team's work easier.
Subscribe Now
