Finance and analysis prompts

Month-end is done, the numbers are in a table, and someone needs the story behind them by tomorrow. Use this to draft the commentary and to find the variances you cannot yet explain.

Act as an FP&A manager who writes variance commentary that explains what happened without guessing.

Budget versus actual table with line items, budget, actual, and variance: [PASTE THE BUDGET VS ACTUAL TABLE]
Period covered: [PERIOD]
What I already know about the drivers (timing, volume, price, one-time items, reclasses): [KNOWN DRIVERS]
Who reads this and how much detail they want: [AUDIENCE]
The threshold below which we do not comment: [MATERIALITY THRESHOLD]

First, recompute every variance in dollars and percent from the table and tell me if any of my figures do not tie or if a line looks misclassified.

Then, for each variance above the threshold, write two to four sentences covering: what moved, the driver if I gave you one, whether it looks like timing or permanent, and the action or watch item that follows. Where I gave no driver, write 'driver not yet identified' and offer the two most likely explanations as questions to investigate, clearly labeled as hypotheses. Never present a guess as a fact.

Open with a three-sentence summary of the period, group the commentary by revenue, cost of sales, operating expenses, and below the line, and close with the items I should raise with budget owners this week. Plain language, no jargon the audience would not use.

Tip: Recompute the variances yourself in the spreadsheet before you trust the narrative; AI arithmetic slips are common and quiet. The hypotheses list is the most useful output, so send those questions to the budget owners before you write another word.

You need money for a hire, a tool, or a project and your first draft is a list of features. Use this to build the case in the terms finance uses to say yes.

Act as a finance business partner who has watched hundreds of budget requests succeed or fail and knows what the approver is really asking.

What I am requesting: [WHAT YOU ARE REQUESTING]
Amount, timing, and whether it is one-time or recurring: [AMOUNT AND TIMING]
The problem it solves and what that problem costs us today, with how I estimated it: [PROBLEM AND CURRENT COST]
Expected return, savings, or risk reduction, and how I estimated it: [EXPECTED RETURN AND BASIS]
What happens if it is not funded: [CONSEQUENCE OF NOT FUNDING]
Who approves and what they care about: [APPROVER AND THEIR PRIORITIES]

Write a one-page justification with: the ask in one sentence; the problem in numbers; the proposed solution; a cost breakdown table; the expected return with assumptions shown, in a conservative case and an expected case; alternatives considered, including doing nothing; risks and how we would notice them early; and the two or three measures we will report to show it worked.

Then list the eight questions a skeptical CFO would ask, each with the honest answer from my inputs or 'need to find out'. Challenge any estimate of mine that looks optimistic and say why. Do not invent benchmarks or industry figures; use only what I gave you.

Tip: The conservative case is what gets approved, so write it such that you would still want the money if that is all you got. Bring the CFO-questions list to the meeting, because they will ask at least five of them.

Department heads, a new manager, or a board member need to understand the numbers well enough to make decisions, and the statement alone is not doing it. Use this to write the walkthrough and the questions they should be asking.

Act as a patient finance teacher who explains statements to smart people who never took accounting.

The income statement or summary, with confidential details removed as needed: [PASTE THE P&L OR SUMMARY]
Who is reading it and what decisions they influence: [AUDIENCE AND THEIR DECISIONS]
What I most want them to understand or do differently: [KEY MESSAGE]

Write a walkthrough that:
1. Explains the statement top to bottom in plain English: what each major line means in this business, with one concrete example per line.
2. Points out the five things a reader should notice, each with the number and why it matters to their decisions.
3. Defines every term used (gross margin, EBITDA, accruals, whatever appears) in one sentence each, in a glossary at the end.
4. Distinguishes what the statement shows from what it does not, such as cash timing, one-off items, and anything I flagged.
5. Suggests four questions this audience should ask finance every month.

Rules: use only the numbers in the statement and recompute any ratio you cite, showing the calculation. Do not speculate about causes I did not give you; phrase them as questions. Keep it under 700 words. This is an explanation of the numbers, not advice on what to do with them.

Tip: Paste a version with rounded numbers or percentages if the real figures are sensitive; the explanation works just as well. Read the glossary first: if the AI defined a term differently from how your company uses it, fix that before anyone else reads it.

You are building or updating a forecast and the assumptions live in your head and a few cell comments. Use this to write them down, rate them, and find the ones that will embarrass you.

Act as a forecasting analyst who believes a forecast is only as good as its written assumptions.

What I am forecasting and the horizon: [WHAT YOU ARE FORECASTING]
Summary of historical performance and trend: [HISTORICAL SUMMARY]
Known changes coming (pricing, headcount, contracts, seasonality, launches, losses): [KNOWN CHANGES]
My draft assumptions, with the number and the reasoning for each: [DRAFT ASSUMPTIONS]
How the forecast will be used and by whom: [HOW IT WILL BE USED]

Build an assumptions register as a table with columns: assumption, value, basis (history, contract, management judgment, or external), source, confidence (high, medium, low), sensitivity (how much the result moves if this is wrong by a plausible amount), owner, and review date.

Then:
1. Challenge each medium- or low-confidence assumption with the strongest reason it could be wrong.
2. Identify the three assumptions that drive most of the outcome and propose a best, base, and downside value for each, with reasoning.
3. Flag any assumption that contradicts the history I gave you or another assumption.
4. List the data I should collect to turn judgments into evidence.

Do not produce forecast totals; I will run the numbers in the model. Do not supply external benchmarks or growth rates I did not give you, and label anything you infer as your own assumption.

Tip: Paste the register into the model as its own tab and link the key cells to it, so a change in an assumption is visible everywhere. When the forecast misses, the register tells you which belief was wrong, which is the whole point.

The board pack is forty pages and the directors will read one. Use this to write the page that tells them what happened, what matters, and what you need from them.

Act as a chief of staff to a CFO who writes board summaries that directors read in five minutes and remember.

This month's financials or KPI table, with comparison to prior period and plan: [PASTE THE FINANCIALS AND COMPARISONS]
Key events this period (wins, losses, hires, incidents, decisions made): [KEY EVENTS]
Risks and what we are doing about them: [RISKS AND RESPONSES]
Decisions or help we need from the board: [DECISIONS NEEDED]
What the board cared most about last meeting: [LAST MEETING CONTEXT]

Write a one-page summary with:
1. Headline: the period in one sentence, honest about direction.
2. Three things that matter, each with the number, the comparison, and one sentence on why.
3. A compact table of the six to eight metrics that matter most: actual, plan, prior period, variance.
4. Risks and responses in three bullets.
5. Asks: exactly what we want the board to decide, with the recommendation and the alternative.
6. A closing line on what to watch next month.

Rules: every number must come from the table I gave you; if you compute anything, show the calculation in a note at the bottom so I can check it. Do not soften bad news or bury it. No jargon a non-finance director would not know. Under 450 words.

Tip: Send the page to one director you trust two days early and ask what they would question. Board packs contain material nonpublic information, so anonymize or use your approved enterprise deployment; never a consumer chatbot.

An invoice landed and something feels off: the total, a line you do not recognize, or a rate that changed. Use this to do a line-by-line match and a fraud check before approval.

Act as an accounts payable analyst who catches billing errors and fraud by checking every invoice against what was agreed and delivered.

Invoice details, line by line, with bank and account details removed: [PASTE THE INVOICE LINES]
The purchase order or the contract's pricing and payment terms: [PASTE THE PO OR CONTRACT TERMS]
What was actually delivered or performed, and when: [WHAT WAS DELIVERED]
Prior invoices from this vendor this period, if any: [PRIOR INVOICES THIS PERIOD]

Produce:
1. A match table: each invoice line against the PO or contract, with quantity, unit price, extended amount, and a status of match, mismatch, or not on the agreement.
2. Math check: recompute subtotals, tax, discounts, and the total, and show your arithmetic.
3. Discrepancies in order of dollar impact, with the question to ask the vendor for each.
4. Fraud and error red flags to consider: duplicate invoice numbers or amounts, changed remittance details, round-number totals, services billed before delivery, terms that differ from the contract, or a rate that changed without an amendment.
5. A short, polite email to the vendor listing the questions.

Do not approve or reject; give me the evidence. If the contract terms I pasted are ambiguous, say so and quote the clause.

Tip: Verify any change in a vendor's bank details by phone, using a number you already had on file, never one on the invoice or the email. That single habit stops most payment-diversion fraud, and no AI check replaces it.

You are about to sign, or you already have, and nobody can say what the contract actually costs over its life. Use this to surface the money terms, the dates to calendar, and the questions for your lawyer.

Act as a commercial analyst who reads contracts for their financial consequences and knows when to hand off to a lawyer.

The contract sections on pricing, payment, term, renewal, termination, and fees or penalties, pasted in full: [PASTE THE CONTRACT SECTIONS]
What I believe we agreed to, in my own words: [WHAT YOU EXPECTED TO AGREE TO]
Expected volume or usage over the term, and the start date: [EXPECTED USAGE AND START DATE]

Give me:
1. A plain-English summary of every financial obligation: price, escalators, minimums, overage rates, payment terms, late fees, and what triggers each.
2. A total-cost table for the full term at expected usage and at usage 25 percent higher and lower, showing your formulas so I can check them.
3. Renewal and termination mechanics: notice windows, auto-renewal, early termination costs, and the exact dates I need to calendar based on the start date.
4. Every place where the contract differs from what I expected, quoting the clause.
5. Ambiguities and risks, ranked, phrased as questions for counsel.

Quote clauses rather than paraphrasing when the wording matters. This is a financial read, not legal advice: do not tell me whether to sign, and flag anything about liability, indemnity, or governing law as a lawyer's question.

Tip: Put the auto-renewal notice deadline in your calendar the day you sign; that clause quietly costs businesses more than any other. And check your company's policy before pasting a counterparty's contract into any AI tool, since many contracts have confidentiality clauses that cover their own terms.

You want to know whether a product, service, or customer segment actually makes money, and you need the model logic more than a number. Use this to get the structure, formulas, and sensitivities you will then build yourself.

Act as a financial modeler who designs simple, transparent models that a non-specialist can rebuild and audit in a spreadsheet.

What I am modeling (a product, a service line, a customer segment, a location): [WHAT YOU ARE MODELING]
Prices and how customers pay: [PRICING AND PAYMENT STRUCTURE]
Cost components, whether each is variable, semi-variable, or fixed, and the amounts I know: [COST COMPONENTS]
Volume assumptions and how customers are acquired and retained: [VOLUME AND RETENTION ASSUMPTIONS]
The decision this model needs to support: [DECISION TO SUPPORT]

Lay out:
1. The model structure as a list of input cells, calculated cells, and outputs, with every formula written in words and then as a spreadsheet formula I can type.
2. The unit economics walk: revenue per unit, variable cost per unit, contribution margin, fixed costs, and break-even volume, computed from my inputs with the arithmetic shown.
3. The three sensitivities that matter most for the decision, with the input to flex and the range to test.
4. The assumptions that carry the most risk and what data would firm them up.
5. Sanity checks: results that would indicate an error in my inputs.

Do not invent costs, rates, or benchmarks I did not give you; leave a clearly marked input cell instead. I will verify every number.

Tip: Build the spreadsheet yourself from the AI's structure rather than asking it for the numbers; the model you built is the one you can defend in a meeting. Then paste your outputs back and ask it to look for errors.

Profit looks fine on paper but the bank balance keeps surprising you. Use this to build a weekly cash view from your receivables, payables, and payroll, and to see where the low point is.

Act as a treasury analyst who builds 13-week cash flow forecasts for small and mid-sized businesses.

Opening cash balance and any credit line available: [OPENING CASH AND CREDIT LINE]
Receivables aging and what I expect to collect and when, with customer names removed: [PASTE RECEIVABLES AND EXPECTED COLLECTIONS]
Payables, payroll dates, rent, loan payments, taxes, and other scheduled outflows: [PASTE SCHEDULED OUTFLOWS]
Expected new sales and their typical collection lag: [EXPECTED SALES AND COLLECTION LAG]
Known one-time items in the next 13 weeks: [ONE-TIME ITEMS]

Build:
1. A week-by-week table: opening balance, receipts by category, disbursements by category, net flow, closing balance, and available liquidity including the credit line.
2. A list of every assumption you made to place a receipt or payment in a week, so I can correct them.
3. The low point: which week, how low, and the three items that most affect it.
4. Timing levers to consider if the low point is uncomfortable, on collections and on payments, in plain terms. Do not recommend borrowing or investment decisions.
5. Spreadsheet formulas for the running balance so I can maintain the forecast weekly.

Show your arithmetic. If any input is missing or inconsistent, ask before building rather than filling the gap with a guess.

Tip: Update it every Monday with actuals from the bank; the forecast gets good after four or five weeks of corrections. Keep customer and employee names out of what you paste, and use your accounting system's exports rather than retyping.