What Utility Accountants Still Need to Know When AI Does the Close

AI tools can now draft month-end accruals, suggest account coding on work orders, and build a first pass at a variance explanation. That part of the job is getting faster.

What has not changed is who signs off. The auditor and the commission staff still hold the utility's accounting staff responsible for the numbers. "The software coded it" is not an answer anyone will accept.

That creates a problem for training. Most of us learned the process by doing it by hand, over and over, until we could spot an entry that looked wrong. If AI does the repetitive work, new staff lose the repetition that built that judgment. The question for every utility finance office is how new people learn the process well enough to review work they never had to do themselves.

Where AI Gets Utility Accounting Wrong

General-purpose AI tools are trained mostly on commercial accounting. Utility accounting under the FERC Uniform System of Accounts has rules that a commercial model does not apply by default. The errors I see most often fall into a few categories.

1. Capital vs. O&M. Replacement of a retirement unit is capitalized. Replacement of a minor item of property is maintenance. An AI tool without the utility's retirement unit list will guess, and it tends to guess O&M.
2. Retirements and cost of removal. When plant is replaced, the original cost of the old unit comes out of Account 101 and is charged to Account 108, along with cost of removal. AI often books the new plant and leaves the old unit on the books, or expenses the removal cost.
3. Plausible account numbers that do not exist. AI will produce an account number that looks right and is not in the Uniform System of Accounts, or is a commercial chart of accounts number.
4. Stale or blended rules. AI may apply FASB treatment where regulatory accounting governs, or rely on guidance that has been superseded.
5. Regulatory context it cannot see. Deferrals under a commission order, CIAC terms in a specific line extension agreement, and company capitalization policies are not in the model. It will not know to ask.

None of these errors is obvious from the entry itself. The debits equal the credits. The descriptions read well. You catch them only if you know what the entry should have been.

The Skills That Still Matter

The reviewer does not need to be faster than the AI. The reviewer needs to know what the right answer looks like. For utility accounting staff, that comes down to process knowledge in a handful of areas.

• Month-end close. AI drafts accruals, reconciliations, and recurring entries. The reviewer has to know which accounts should move each month and by roughly how much.
• Work orders and the CPR. AI suggests account coding and unitization. The reviewer has to know retirement units, cost of removal, salvage, and AFUDC eligibility.
• Construction and CIAC. AI drafts cost summaries and CIAC entries. The reviewer has to know the line extension terms and how CIAC is recorded.
• Revenue requirement and rates. AI builds spreadsheet models and allocators. The reviewer has to know whether the allocators and test-year adjustments make sense.
• Regulatory reporting. AI drafts FERC Form 1 notes and variance narratives. The reviewer has to tie them to the general ledger and to prior-year filings.

The common thread is that the reviewer has to know the process before the AI output is useful. Without that, review becomes a read-through, not a check.

Worked Example: An AI-Drafted Pole Replacement

A crew replaces a damaged distribution pole. Figures are illustrative.

• New pole, installed: $4,200 (materials $1,900, labor and overheads $2,300)
• Cost of removing the old pole: $600
• Original cost of the old pole on the CPR: $950
• No salvage

The AI tool produced this entry from the work order:

• Dr 107 Construction Work in Progress — 4,200
• Dr 593 Maintenance of Overhead Lines — 600
• Cr 154 Materials and Supplies — 1,900
• Cr Payroll and overhead clearing — 2,900

It balances, and each line has a reasonable description. It has two errors.

1. Cost of removal was expensed to Account 593. It should be charged to Account 108, Accumulated Provision for Depreciation (through retirement work in progress).
2. The old pole was never retired. Its $950 original cost stays in Account 364, Poles, Towers, and Fixtures, and keeps depreciating.

The corrected entries:

• Dr 107 Construction Work in Progress — 4,200
• Dr 108 Accumulated Provision for Depreciation (cost of removal) — 600
• Cr 154 Materials and Supplies — 1,900
• Cr Payroll and overhead clearing — 2,900

• Dr 108 Accumulated Provision for Depreciation (retirement) — 950
• Cr 101 Electric Plant in Service, Account 364 — 950

The effect of the uncorrected entry: O&M overstated by $600 in the current year, and plant in service overstated by $950 until someone finds it. On one pole that is small. Across a storm restoration with several hundred poles, it moves operating income and the depreciation base.

A reviewer catches this only by knowing that replacement of a retirement unit triggers a retirement, and where cost of removal belongs. The AI tool will not flag either one.

What This Means for Training New Staff

If AI is going to do the first draft, training has to change in three ways.

1. Teach the process first, then the tool. New staff should work through the manual version of a close, a work order, or a retirement before they review an AI version of it. They do not need to do it for years. They need to do it enough to know what right looks like.
2. Practice review, not just preparation. Give staff AI-drafted entries with known errors and have them find the errors. That is the skill the job now requires, and it can be taught directly.
3. Practice the conversations. Some firms are now using AI role-play to train new staff on how to request information and push back professionally. The same approach works in a utility office: asking engineering for work order detail, explaining a variance to management, or responding to a data request from commission staff.

Our courses at UtilityEducation.com are adding AI review exercises built on this approach, starting with the construction and work order material. The goal is the same as it has always been: staff who understand the accounting well enough to stand behind the numbers, whoever or whatever prepared them.