Implementing AI in the Utility Finance Office Without Displacing Your Staff

In an earlier article I suggested starting AI implementation with a task-level inventory: list what each position actually does in a month, how many hours each task takes, and sort the tasks into three categories. Mechanical and rules-based. Judgment-based but pattern-driven. Judgment-based and context-dependent.

The next question is what to do with the hours AI frees up in the first category. The default assumption in a lot of industries is that those hours become fewer positions. In a utility or co-op finance office, I don't think that is the right answer, and in most offices it isn't the necessary one.

Why Displacement Is Not the Default Here

Most utility finance offices are already lean. A distribution co-op or municipal utility may run the entire accounting function with four to eight people. There is not a lot of slack to cut.

The workload is also going up, not down. Large load agreements, CIAC and contribution questions, more frequent rate filings, new GASB and FASB standards, and board and commission requests for more analysis all add hours. Most offices I've worked with have a list of work that never gets done: the depreciation study that is overdue, the cost of service update, the CPR cleanup, the reconciliations that get done quarterly instead of monthly.

And retirements are coming. The people who know how the work order system ties to CWIP, or where the numbers in the last rate case came from, are close to the end of their careers. Replacing them is slow and expensive.

Put those together and the math changes. The hours AI frees up are not surplus. They are the hours you need to cover growing work and a shrinking pool of experienced staff.

Redesign the Role Before You Install the Tool

The order matters. If you install the tool first, the freed hours show up as idle time and someone will eventually ask why the position is still needed. If you redesign the role first, the freed hours already have a place to go.

Here is how that works for one position, using illustrative numbers. A staff accountant works about 160 hours a month. The task inventory shows roughly 50 of those hours in category one:

• Invoice entry and coding: 24 hours now, 6 hours with AI (review time only)
• Bank and subsidiary ledger reconciliations: 10 hours now, 3 hours with AI
• Pulling and formatting data for monthly reports: 16 hours now, 4 hours with AI
• Total: 50 hours now, 13 hours with AI

The AI does not take the full 50 hours. Someone still has to review what it produced, and that review time is real. In this example about 37 hours a month move out of category one.

Those 37 hours go to work the office needs and is not getting done today:

1. Monthly reconciliation of the CPR to the general ledger instead of annual
2. Work order closing reviews, catching unitization and retirement errors before they sit for a year
3. Variance analysis that explains why an account moved, not just that it did
4. Supporting the next rate filing or cost of service update
5. Learning the job of the person who is retiring in three years

The position still exists. It spends less time keying data and more time on categories two and three, which is where the utility gets the most value from an experienced accountant anyway.

Write the redesigned role down. Update the job description with the new task mix before the tool goes live, so the change is a decision management made, not something that happened by default.

Use Attrition, Not Layoffs, to Adjust Headcount

In some offices the inventory will show more freed hours than there is new work to absorb them. Even then, layoffs are rarely the only option.

Compare the timeline for AI implementation with your retirement schedule. Most finance offices have one or two people within five years of retirement. If the freed hours in the office add up to roughly one position, the practical answer is often to not refill a position when someone retires, and to use the intervening years to move that person's knowledge to the staff who remain.

That is slower than a reduction in force. It also keeps the institutional knowledge in the building, avoids the morale damage that follows a layoff, and doesn't signal to the rest of the staff that working with AI is how they lose their job.

Retrain for Review, Not Just for the Tool

Most AI training focuses on how to use the tool: how to write a prompt, where to click. That part is quick to learn.

The harder part is teaching staff to review what the tool produces. A reviewer has to know what the right answer looks like: which accounts should move each month, when a replacement triggers a retirement, whether an allocator in a rate model makes sense. That comes from knowing the underlying process. Staff who move from data entry to review need that process training, and it should come before or alongside the tool, not after.

Keep the Internal Controls in Place

AI does not change who is accountable for the numbers. Before the tool goes live, decide:

1. Who reviews AI-prepared entries, and at what dollar threshold a second reviewer is required
2. How the review is documented, so the auditors can see it happened
3. Which tasks AI is not allowed to finalize on its own, such as anything posting to plant accounts or regulatory assets

The people reviewing AI output are part of the control environment. That is another reason to keep experienced staff in place.

Tell the Staff What Is Happening, Early

People assume the worst when they see a new tool and hear nothing about what it means for them. Tell staff at the start what the plan is: which tasks are moving to AI, what their role will look like afterward, and whether any positions will change through attrition. Involve them in building the task inventory. They know where the hours actually go better than management does.

Bottom Line

In a utility finance office, AI is most useful as a way to cover growing work and retirements with the staff you already have. That outcome takes planning: redesign the roles first, use attrition instead of layoffs, train for review, keep the controls, and tell the staff early. Offices that skip those steps risk losing the experienced staff they need to review what the tool produces.