|
BUILD WITH AI |
BUILD WITH AI · PART 7 OF 17
Build a month-end flux commentary workflow with an AI assistant
The draft takes minutes; the check still needs a human who knows why the numbers moved.
Month-end flux commentary is one of those tasks that takes longer to format than to think through. A controller or senior accountant pulls the trial balance comparison, identifies the accounts with material moves, and then writes three to five sentences per line explaining what happened. The writing itself is mechanical once you know the story, which makes it a reasonable candidate for a first draft from a general-purpose AI assistant.
BUILD WITH AI
Month-end flux commentary, step by step
This workflow turns a two-period trial balance export into a first-draft flux commentary in one assistant session. The output is a structured draft with bracketed placeholders wherever the assistant cannot determine cause from the data alone.
1 |
Export a two-period trial balance from your accounting system as a CSV, with columns: account code, account name, prior period balance, current period balance, and variance amount. Remove subtotal rows. |
2 |
Open a general-purpose AI assistant that accepts file or text input, and paste or upload the CSV. |
3 |
Set your materiality threshold by including this instruction at the top of your prompt: 'Only comment on accounts where the absolute variance exceeds [your dollar threshold].' |
4 |
Paste the following prompt: 'You are preparing month-end flux commentary for an internal workpaper. The attached trial balance shows two periods. For each account where the absolute variance exceeds the threshold I specified, write two to three sentences explaining the likely cause based on the account name and the direction and size of the movement. Where the cause cannot be reasonably inferred from the account name alone, write the two sentences you can support and add a bracketed note: [Cause unconfirmed: reviewer to verify against subledger]. Group accounts by: revenue, cost of goods sold, operating expenses, other income and expense, balance sheet. Do not invent subledger detail. Do not remove any bracketed note to make the commentary appear more complete.' |
5 |
Review the draft output and confirm that every account meeting your threshold has an entry and that no account below the threshold has been included. |
6 |
Resolve every bracketed placeholder by pulling the relevant subledger report, accrual schedule or supporting document and writing the confirmed cause in plain language. |
7 |
Trace each figure the assistant states (dollar amounts, percentages, period references) back to the trial balance CSV before the commentary is copied into the workpaper. |
8 |
Have the responsible accountant or controller sign off on the final commentary as a named preparer step in the workpaper, not as an AI-generated exhibit. |
The workflow below exports a two-period trial balance, feeds it to an AI assistant with a structured prompt, and returns a draft commentary organized by account group. The assistant flags accounts where the variance exceeds a threshold you set, proposes a plain-language explanation based on the labels and amounts, and leaves a bracketed placeholder wherever the cause is ambiguous. That placeholder is the most useful output: it tells the reviewer exactly where human judgment is still required rather than hiding a gap behind confident prose.
Where this workflow breaks down is predictable. The assistant cannot see the subledger, the invoice detail, or the accrual schedule behind the numbers. It will sometimes assign a plausible but wrong narrative to a large variance, particularly in accounts like prepaid expenses or intercompany eliminations where the label alone does not explain the movement. A reviewer who knows the period can catch those in a pass; a reviewer who is seeing the numbers for the first time may not. The workflow is a time-saver for experienced preparers, not a substitute for account-level knowledge.
The finished commentary should be compared line by line against the trial balance export before it goes into any report or workpaper. Any figure the assistant states should be traced back to the source file. Bracketed placeholders must be resolved, not deleted. If the draft looks too clean, that is a signal to slow down: an AI assistant that hedges less than expected is often one that has papered over a gap rather than surfaced it.
WORKED EXAMPLE
In practice
A controller at a 12-entity services firm is closing the month just ended. She has a two-period trial balance for one entity as a CSV: 47 accounts, prior and current period balances, variance column already calculated. She needs flux commentary for the management report due in two days.
What came back. The assistant returned commentary for 19 accounts, grouped correctly. Most revenue and payroll entries had clean two-sentence explanations consistent with the account labels and variance directions. Four accounts in prepaid expenses and intercompany receivables carried bracketed placeholders. One entry for a large increase in professional fees stated a cause (outside counsel costs) that was plausible but not supportable from the account name alone, and the controller added a placeholder there herself before moving on.
How it was checked. The controller opened the CSV alongside the draft and confirmed that every dollar figure the assistant cited matched the variance column exactly, and that no account below $10,000 had been included.
A constructed example. The prompt is usable as written; the figures show the shape of a result, not a measured one.
WHEN TO USE IT
| WHEN NOT TO
|
WHAT TO TAKE FROM THIS
| Set a variance threshold in the prompt; commentary on every account creates noise. | |
| Bracketed placeholders in the draft are features, not errors: resolve each one before filing. | |
| Trace every figure the assistant states back to the trial balance export before the workpaper is signed. |
Finally, a Monday that doesn’t feel like one.
Imagine a Saturday where you're not catching up on the CRM you didn't touch all week.
HubSpot's new self-updating CRM doesn't wait for you to log back in. Friday's calls are already logged. The deal that needs a nudge Monday morning is already flagged. The automation you set up once keeps running without you thinking about it. None of it needed you over the weekend.
Customers are already seeing it pay off: 45% more deals closed. Finally, you start Monday caught up, not behind.
FROM OUR PARTNERS
QUESTIONS THIS ANSWERS
Can an AI assistant write month-end variance commentary without human review?
No. The assistant works from account labels and amounts only. It cannot see subledger detail, accrual schedules or the business context behind a variance. A reviewer with knowledge of the period must check every line before the commentary is filed.
What file format works best for the trial balance export?
A CSV or plain-text export with columns for account code, account name, prior period balance, current period balance, and the variance amount is sufficient. Remove any subtotal or header rows that could confuse the assistant's column parsing.
Does this workflow require a specific AI product?
No. Any general-purpose AI assistant that accepts a text or file input and returns structured prose will work. The quality of the draft depends more on the prompt and the cleanliness of the export than on the specific tool.
SOURCES
Where this comes from
The career portal does the rest.
GO DEEPER
Go deeper
IN THIS SERIES
Previously: First-pass GL codes from one prompt
Next: Build with AI: accrual review (coming)
An original workflow written for practitioners. Replicate it in a sandbox first; nothing here replaces your review.
How accountants are using AI, automation and smarter workflows to close faster, audit cleaner, and free up time for real work.
Accounting Stack · Audit Friendly Data · Accounting & Finance Jobs
Audit Friendly · modernaccounting.ai

