Your month-end close probably looks like this: a stack of acts, a bank statement, a supplier reconciliation report, and three hours of line-by-line matching in Excel. One missed line — and the VAT return is off. The good news: large language models can now do the boring part — match counterparties, flag discrepancies in amounts and dates, and draft explanations for your accountant. The catch: only if you use the right prompts and know when manual reconciliation is still faster. This roundup gives you battle-tested prompts for accounting and finance, plus a clear criterion: use them when you process more than ~50 documents a month and are ready to anonymize data before sending it to the cloud.
How many documents before prompts beat manual reconciliation?
Manual reconciliation is fine up to about 50 primary documents per month. Beyond that, the time sink grows non-linearly: every new counterparty adds cross-checks. LLM prompts shine when you have 100+ acts, invoices, and payment orders — they cut matching time from hours to minutes, but require anonymization (replace names and amounts with placeholders). If you handle fewer than 50 documents, stick to Excel — the setup cost of prompts won't pay off. More on the topic: “Excel on Autopilot: 12 Prompts That Turn an LLM Into Your Spreadsheet Analyst — From Pivot Tables and Formulas to Cleaning Dirty Data”.
Prompt 1: Reconciliation of acts and invoices — find mismatches in amounts and dates
This prompt turns a messy list of acts and a bank statement into a discrepancy report. Use it when you have two CSV exports: one from your accounting system (acts) and one from the bank (payments). The LLM matches by counterparty and amount, then flags differences in dates or sums. More on the topic: “Accounting and Reporting: The Accounting Law, Standards and Journal Entries — How to Navigate a Typical Scenario”.
Prompt:
You are a reconciliation assistant. I will provide two tables: Acts (columns: counterparty, act_number, act_date, amount) and Payments (columns: counterparty, payment_date, amount, purpose). Match each act to a payment by counterparty and amount. Output a table with columns: counterparty, act_number, act_amount, payment_date, payment_amount, status (matched / amount mismatch / date mismatch / missing payment). For mismatches, add a comment explaining the difference. Do not invent data — if a field is missing, write "N/A".
Example: you upload 20 acts and 18 payments. The LLM returns 2 acts without payments and 1 payment with a 10% higher amount — you instantly see where to dig.
Prompt 2: Draft explanations for your accountant
When discrepancies are found, your accountant needs a short note: what's wrong, which document, and what to check. This prompt generates a draft explanation in plain language, ready to paste into an email or chat.
Prompt:
Based on the discrepancy report below, write a short explanation for the accountant. For each mismatch, state: counterparty, document number, the issue (amount or date), and a suggested action (request a corrected act, check payment purpose, etc.). Keep it under 150 words per issue. Use a professional but friendly tone. More on the topic: “Profession: Accountant from Source Documents to Reporting — The Complete Path into Accounting: Areas, Primary Documents, Taxes, Payroll, Reporting, 1C, and AI Tools”.
Example output: "Act #45 from Supplier A shows 1,200 USD, but payment on 05.11 is 1,320 USD. Please check if VAT was included or if there's a typo in the act."
Prompt 3: Anonymize data before sending to the cloud
Before you paste anything into ChatGPT or another cloud LLM, replace sensitive data. This prompt helps you create a mapping table and a sanitized version of your document.
Prompt:
I will give you a list of counterparties and amounts. Replace each counterparty name with "Counterparty_1", "Counterparty_2", etc., and each amount with a placeholder like "AMOUNT_1". Keep the structure intact. Output two tables: the anonymized document and a mapping table (original → placeholder). Do not include any real names or numbers in the anonymized version.
Example: you paste a list of 10 suppliers; the LLM returns a safe version you can use in prompts, plus a mapping to restore real names in the final report.
Prompt 4: Summarize the month's financials for a non-financial founder
Founders often don't need a full P&L — they need a 5-bullet summary: cash position, top expenses, overdue receivables, and one red flag. This prompt turns a CSV export into a plain-language brief. More on the topic: “Spreadsheets Don't Lie, They Just Return #N/A: How to Make an LLM Your Formula, Pivot and Apps Script Co-Pilot”.
Prompt:
You are a financial analyst. Given a CSV with columns: date, category, amount, type (income/expense), produce a summary for a founder: 1) total income and expenses for the month, 2) top 3 expense categories, 3) any overdue receivables (if data available), 4) one potential risk you notice. Use simple language, no jargon. Max 200 words. More on the topic: “Profession: Financial Manager from Accounting to CFO — The Complete Path in Corporate Finance: Where It Breaks and How to Avoid Mistakes”.
Example: the LLM reports "Income 12,000 USD, expenses 9,500 USD; top expenses: rent (3,000), salaries (4,000), software (800). Risk: two invoices marked 'unpaid' for over 30 days." More on the topic: “Medical Expense Tax Deduction: How to Avoid a Rejection and Get Your Money Back”.
When prompts fail: the data quality trap
LLMs are only as good as your input. If your acts have inconsistent formatting (e.g., dates as "05.11.2026" vs "2026-11-05"), the model may misread them. Always normalize dates and amounts before prompting. Also, LLMs can't verify mathematical correctness beyond simple sums — for complex VAT calculations, use Excel formulas and let the LLM only draft explanations. More on the topic: “Your Excel Chart Is Not an Analysis: LLM Prompts That Hunt Anomalies Before You Draw Anything”.
Bottom line
Use these prompts when you have more than 50 primary documents per month and are comfortable anonymizing data. Start with Prompt 1 for reconciliation, then add Prompt 2 for explanations. If you're below 50 documents, manual Excel reconciliation is still your friend. Try one prompt this week — pick your messiest reconciliation and see how much time you save.