Month-end close used to be the worst week of my month.
Three days of categorizing transactions. Half a day of bank reconciliations. Another day pulling together variance analysis and management reports. By Friday I was exhausted and behind on everything else.
Now month-end close takes me a day, sometimes less, depending on client complexity. Same accuracy. Same documentation. Same review steps. Just AI doing the routine work and me doing the judgment work.
This is the complete workflow I use, broken down step by step, with the actual prompts I run at each stage.
Before I run a single prompt, I do three things that pay off across the entire close:
1. Build a client profile document. One page per client with their chart of accounts, common vendors, owner names (anonymized), and any quirks ("rent paid quarterly, not monthly"). I paste this into AI prompts to give context. Saves 30 minutes of correcting AI output that didn't know the client.
2. Standardize your data exports. Every accounting software exports transactions slightly differently. Spend an hour creating a clean CSV template and use it for all clients. AI handles consistent formats much better than messy ones.
3. Save your prompts. The prompts in this article are starting points. After you customize them for your practice, save them in a Notion doc or a Google Doc you can paste from instantly. Don't rewrite from scratch every month.
This is where AI saves the most time across the entire close. For most clients, transaction categorization is 50% of the close work. AI cuts it by 70-80%.
"You are an experienced senior bookkeeper. I will paste a list of bank transactions for [CLIENT NAME, ANONYMIZED] for [MONTH]. For each transaction, provide:
1. Suggested account category from this list: [PASTE CHART OF ACCOUNTS]
2. Confidence level (High/Medium/Low)
3. Brief reasoning
4. Flag for any transactions that need human review
Format as a table. Be conservative with High confidence — only mark items where the categorization is unambiguous.
Client context: [PASTE CLIENT PROFILE]
Transactions: [PASTE TRANSACTIONS]"
You'll get a categorized table. Don't trust it blindly — use the confidence levels to triage your review.
Before posting, scan the categorized output for these patterns:
For batches over 100 transactions, run them in chunks of 50. Accuracy drops noticeably above 50 per prompt.
AI can't actually reconcile bank statements for you — that requires access to your accounting software. But it can dramatically speed up the matching process and the investigation of differences.
"I have two lists for [MONTH]: bank statement transactions and general ledger entries. Compare them and:
1. Match transactions that correspond to each other (same date or close, same amount)
2. Identify unmatched bank transactions (could be outstanding deposits/checks)
3. Identify unmatched ledger entries
4. Flag any matches with amount discrepancies
5. Note any unusual patterns (duplicates, voids, etc.)
Format as four tables: Matched, Unmatched Bank, Unmatched Ledger, Discrepancies.
Bank transactions: [PASTE]
Ledger entries: [PASTE]"
The matched table is your starting point — verify a sample to make sure AI got it right, then move on.
The unmatched lists are where you spend your time. For each item:
Always reconcile the totals manually. AI handles the matching well but occasionally misses items in long lists. The total reconciliation is your safety net — if the totals match, the detailed work is probably right.
For each accrual you typically book, AI can prepare the journal entry from your description.
"Prepare month-end accrual journal entries for [CLIENT NAME] for [MONTH]. Use these accounts: [LIST RELEVANT ACCOUNTS WITH NUMBERS].
Accruals to book:
1. [DESCRIBE ACCRUAL — e.g., "Monthly rent of $5,000 paid in arrears"]
2. [DESCRIBE ACCRUAL]
3. [Continue for each]
For each, provide: Date, Account, Debit, Credit, Description.
Format as standard journal entries. Include applicable tax entries where relevant. Flag anything where you're uncertain about the correct treatment."
Review each entry before posting. AI gets the basic accruals right but occasionally reverses debits and credits, especially for complex entries (deferred revenue, prepaid expenses, etc.).
For new accrual types you haven't booked before, double-check with your accounting reference. This is exactly the kind of nuance AI can get wrong.
This is where AI really shines for clients who require management reporting. Variance analysis is structured work — perfect for AI.
"Analyze the following budget vs. actual results for [CLIENT NAME] for [MONTH] and explain significant variances.
Threshold: flag anything over [5%] OR [$X] variance, whichever is greater.
For each significant variance:
1. Account/category
2. Budget amount vs actual
3. Variance ($ and %)
4. Likely explanation (based on context provided)
5. Whether it's a one-time or recurring issue
6. Recommended action
Group variances into 'favorable' and 'unfavorable' for clearer presentation.
Client context: [INDUSTRY, BUSINESS MODEL, ANY KNOWN EVENTS THIS MONTH]
Budget vs Actual: [PASTE DATA]"
The numerical analysis (variance amounts, percentages) is reliable. AI is good at math when the math is structured.
The "likely explanation" column is your starting point, not the final answer. AI is guessing based on patterns. You add the client-specific context that explains what actually happened.
Verify the variance calculations independently for at least the top 3 items. If those check out, the rest of the math is probably right.
For the explanations, ask yourself: does this reflect what I actually know about this client's month? If not, rewrite it.
By this point, AI has done the grunt work. Now it gets to do the writing.
"Based on the financial data below, prepare a professional monthly financial report for [CLIENT NAME], a [BUSINESS DESCRIPTION] with approximately [REVENUE RANGE] in monthly revenue.
Include:
1. Executive summary (3-4 sentences)
2. Revenue analysis (vs prior month and vs budget)
3. Expense analysis (top categories, significant changes)
4. Profitability metrics (gross margin, net margin, trends)
5. Cash flow summary
6. 3 key findings
7. 2 recommendations
Tone: professional, concise, suitable for a CEO who is not financially trained.
Length: 1-2 pages.
Financial data: [PASTE FINANCIAL DATA]
Variance analysis from prior step: [PASTE]"
You'll get a polished first draft. Spend 15-20 minutes editing:
The result: a report that reads like you wrote it, in 20 minutes instead of 2 hours.
This step doesn't change with AI. Final review is your professional responsibility, and AI doesn't replace it.
What you should review every month:
Document that you performed this review. If your firm requires sign-off, make sure it happens.
For a typical small-business client, here's the time comparison:
Old workflow:
AI-supported workflow:
That's 7+ hours saved per client per month. For a bookkeeper with 5 clients, that's 35+ hours a month — basically a full work week back.
I see accountants make the same two mistakes when implementing AI in their close:
Mistake 1: Trying to automate everything at once. Don't. Pick one step. Master it. Add the next. The accountants who go all-in on AI in one month usually give up by month two because something didn't work and they can't tell which step caused the problem.
Mistake 2: Skipping the review steps to save more time. The verification steps are what make this workflow safe. Skip them and you'll eventually post incorrect entries that take longer to fix than the time you saved.
Start with Step 1 (transaction categorization). Use it for one month with one client. Once you trust the workflow, add Step 2. Build up gradually.
AI doesn't make month-end close easier. It makes it different.
You'll spend less time on data entry and more time on review and judgment. Less time on first drafts and more time on edits. Less time on the routine and more time on the exceptions.
This is, I think, the right trade-off. The routine work was the part of the close that didn't require my expertise. The judgment work is exactly what clients pay for. AI lets me spend more time on the latter and less on the former.
Same accuracy. Same documentation. Half the time.
That's worth the work to set up.
The exact prompts in this workflow (and 47 more) are in the free PDF — including the Categorization Playbook with all the customization tips. Download it here.
Get 50 copy-paste prompts delivered to your inbox — free.
Get the Free PDF →