Short answerThe best LLM prompts for finance give the model a role, your actual numbers, a clear task and a required output format — then ask it to show its work. Group them by job: FP&A and variance analysis, cash and runway, board and investor communications, and close and operations. Always demand the calculation logic and check every number yourself.
A Stanford and MIT study reported by the Journal of Accountancy (2025) found that accountants using generative AI reclaimed about 3.5 hours a week — but the ones who got that lift were not typing "help me with finance." They were giving the model tight, structured instructions. That is the entire difference between an AI assistant that writes a vague essay and one that hands you a board-ready variance table. Below are 18 LLM prompts for finance you can copy, paste and adapt — written for FP&A work, not marketing fluff — grouped by the job they do.
Two ground rules before you start. First, these are written for a generic AI assistant or large language model; use whichever your company has approved. Second, replace every [bracket] with your real numbers, and read the safety note at the end before anything touches a board deck.
What are the best LLM prompts for finance and FP&A work with an AI assistant?
The pattern under every good prompt is the same four parts: role, data, task, format. Tell the model who to be, give it your actual figures, say precisely what to produce, and specify the shape of the answer. Then add one line — "show the calculation behind each number" — so you can verify it. Skip any of these and you get a confident, generic, unverifiable answer, which in finance is worse than no answer.
Group 1: FP&A and variance analysis
1. Budget-vs-actual variance.
"Act as an FP&A analyst. Here is my budget and actuals for [month]: [paste figures by line item]. Produce a variance table with budget, actual, dollar variance and percent variance per line. Then write 4 bullet points explaining the three largest variances. Show the calculation for each variance."
2. Variance commentary in plain English.
"Rewrite this variance table as a short commentary a non-finance founder can read in 60 seconds: [paste table]. Lead with the single most important takeaway. No jargon."
3. Trend and driver analysis.
"Here is [metric] by month for the last 12 months: [paste series]. Identify the trend, call out any month that breaks it, and list the two most likely drivers I should investigate. Flag anything that looks like a data error."
4. Departmental cost review.
"Here is spend by department for [period] vs the prior period: [paste]. Rank departments by percent change, highlight anything growing faster than revenue (revenue growth was [X]%), and suggest three questions I should ask each department head."
5. Rolling forecast reforecast.
"Act as an FP&A analyst. Given actuals for [months elapsed]: [paste], and these assumptions for the rest of the year: [paste], produce a reforecast for the full year by quarter. State every assumption you used and mark which figures are actuals vs forecast."
Group 2: Cash and runway
6. Runway calculation.
"Cash on hand is [$X]. Net monthly burn for the last three months was [$A], [$B], [$C]. Calculate current runway in months using the three-month average burn, then show it again using last month's burn only. State which is more conservative and why."
7. Runway scenarios.
"Using cash of [$X] and monthly burn of [$Y], model runway under three cases: base (burn flat), downside (burn +20%, revenue -15%), and stretch (burn -10%). Give runway in months for each and a one-line summary of what each case implies."
8. Cash-flow bridge.
"Explain the change in cash from [$opening] to [$closing] over [period]. Build a bridge: operating cash, investing, financing and any one-offs, using these figures: [paste]. Flag the single biggest swing."
9. Collections prioritization.
"Here is my aged receivables list: [paste customer, amount, days overdue]. Prioritize which accounts to chase first by dollar value and lateness, and draft a short, firm-but-polite reminder email for the top three."
10. Burn-multiple check.
"Net new ARR this quarter was [$X] and net cash burned was [$Y]. Calculate the burn multiple, explain in one sentence what it says about capital efficiency, and note the rough threshold investors consider healthy."
For deeper cash work, pair these with the approach in AI cash flow forecasting.
Group 3: Board and investor communications
11. Board one-pager.
"Act as a startup CFO. Turn these numbers into a one-page board summary: [paste revenue, growth, burn, runway, key wins, key risks]. Structure: headline metrics, what changed vs last month, top 3 risks, top 3 asks. Keep it under 250 words. Do not invent any number I did not give you."
12. Investor update email.
"Draft a monthly investor update from these inputs: [paste highlights, metrics, lowlights, asks]. Tone: candid and concise. Include a 'help wanted' section. Flag anywhere I have left a gap you would expect an investor to ask about."
13. Metric narrative.
"Write two sentences of narrative for each of these KPIs for a board deck: [paste metric: value, prior value]. State the direction, the magnitude, and the likely cause. If a metric moved in a bad direction, say so plainly."
14. Anticipate board questions.
"Given this board summary: [paste], list the eight hardest questions a sharp board member is likely to ask, and a one-line honest answer to each. Mark any question I currently cannot answer with the data I have."
15. Fundraise metric check.
"Here are my key metrics: [paste ARR, growth, NRR, gross margin, CAC payback, burn multiple]. For each, state whether it is above, at, or below what an investor at [seed/Series A] typically expects, and which one is my weakest story."
Prompt 15 pairs naturally with the SaaS metrics investors actually read.
Group 4: Close and operations
16. Close checklist generator.
"Act as a controller. Build a month-end close checklist for a [SaaS/e-commerce] startup with [team size]. Group by area (cash, AR, AP, payroll, accruals, revenue), and mark which steps a junior can own vs which need my review."
17. Journal-entry explainer.
"Explain, in plain English, what this journal entry does and why it might be posted: [paste entry]. Note anything unusual about it and one question I should ask before approving it."
18. Reconciliation triage.
"Here is a list of unreconciled items: [paste]. Group them by likely cause (timing difference, missing entry, duplicate, error), and tell me which to investigate first and which are probably benign."
The full set maps onto the layers we describe in the full AI CFO stack.
How to use these safely
The prompts are only half the job. The other half is not shipping a wrong number.
- Paste real data only into an approved tool. Never put confidential financials into a random consumer chatbot; use whatever your company has cleared, with data terms you have actually read.
- Always demand the math. Every prompt above asks the model to show its calculation for a reason — so you can check it.
- Verify before it leaves the building. Language models are fluent even when they are wrong. Deloitte (2025) has warned that AI-generated claims and numbers must be grounded in verified source documents before they inform a decision — advice that came after real firms shipped fabricated figures. Spot-check every number against your source of truth. We go deeper on this in where AI hallucinations bite finance teams.
- Treat the output as a junior's first draft. Fast and useful, not final. You own the sign-off.
What this means for you
These 18 prompts turn an AI assistant from a novelty into a genuine FP&A accelerator — the kind of lift that gave the accountants in the Stanford–MIT study back several hours a week. But the lift only shows up if you feed the model real numbers, force it to show its work, and check the output before it reaches a board or an investor.
The one rule to carry into every prompt: the AI writes the draft, you own the number. Speed on the drafting, discipline on the verification. Do both and you get board-ready output in a fraction of the time — without ever forwarding a figure you have not personally confirmed.
Frequently asked questions
What makes a good finance prompt different from a generic one?
Four things: a role ('act as an FP&A analyst'), your real numbers pasted in, a specific task, and a required output format. Generic prompts get generic essays. Finance prompts get usable tables and commentary because you constrain them. Add 'show the calculation for each figure' so you can check the math — an AI assistant is fluent even when it is wrong, so verifiable output matters more than polished output.
Can I paste my company's financial data into an AI assistant?
Only into a tool your company has approved, with data terms you have read. Never paste confidential financials into a random consumer chatbot. Many finance teams use a business-tier or self-hosted assistant precisely so their numbers are not used for training. Check your data-handling policy first; if you do not have one, that is the thing to fix before you scale prompting.
Will an AI assistant do my variance analysis correctly?
It will draft it well and occasionally get a number wrong. Language models predict plausible text, so they can transpose figures or invent a total that looks right. Always ask the prompt to show its calculation, then spot-check the arithmetic against your source data. Use it to write the commentary and structure fast — but you own the final numbers and the sign-off.
Are these prompts safe to use for board and investor materials?
The prompts are safe; unreviewed output is not. Board packs and investor updates are read by people who make decisions about your company, so every figure must be checked against your source of truth before it ships. Treat the AI draft as a first pass a junior analyst handed you: useful, fast, and not to be forwarded until you have verified it line by line.
Sources
- Journal of Accountancy (AICPA & CIMA) — Calculating AI's impact on CPAs: New study quantifies time savings, 2025. https://www.journalofaccountancy.com/news/2025/aug/calculating-ais-impact-on-cpas-new-study-quantifies-time-savings/
- Deloitte — AI doesn't lie, it hallucinates and M&A due diligence must address that, 2025. https://www.deloitte.com/ch/en/services/consulting/perspectives/ai-hallucinations-new-risk-m-a.html
