What are the best AI prompts for financial modeling?

Give the model your own assumptions, historicals and layout, and ask it to build and stress-test a structure, not to supply data it does not have. Keep inputs separate from formulas, then reconcile every total yourself before relying on any output.

Can AI pull live financial data for my model?

Only if you connect it to a data source or paste the figures in. A chat model on its own has no live feed, and a script that claims to fetch prices will fail or invent values. Supply the numbers, and name the source and date.

How do I make an AI-built spreadsheet model easy to audit?

Ask for one inputs sheet, simple formulas that each do one job, and a checks block such as balance sheet balances and cash ties to the cash flow statement. The FAST Standard asks for simple, clear formulas and consistent layout for the same reason.

Best AI Prompts for Financial Modeling (2026)

Prompts for financial modeling#

Five starting points#

Full briefFinancial modeller

Act as a financial modeller. Build a three-statement model (income statement, balance sheet, cash flow) in [Excel / Google Sheets / Python with pandas] for [DESCRIBE YOUR COMPANY AND BUSINESS MODEL]. Use only the historicals and assumptions I give you below, and list any assumption you had to add yourself instead of silently choosing a value. Put all inputs on one sheet, keep each formula to one job, add base, upside and downside scenarios selected by one switch, and add checks that the balance sheet balances and cash ties to the cash flow. [PASTE YOUR HISTORICAL FIGURES AND ASSUMPTIONS]

AdvancedValuation analyst

Act as a valuation analyst. Using the forecast free cash flows, discount rate and terminal growth rate I provide, build a discounted cash flow calculation and show the full arithmetic for each year. Then produce a sensitivity table of value against discount rate and terminal growth, and state what share of the total value comes from the terminal value. Do not look up or invent market data; flag any input I left out. [PASTE YOUR FORECAST CASH FLOWS AND RATES]

AdvancedStartup CFO

Act as a startup CFO. Build a monthly SaaS forecast from my starting figures: new customers, price, monthly churn, expansion, acquisition cost and gross margin. Calculate MRR, ARR, CAC, LTV and payback period, show each formula, and tell me which two assumptions move the result most. [PASTE YOUR STARTING MRR, CUSTOMER COUNT, CHURN, CAC AND MARGIN]

SpecializedQuantitative analyst

Act as a quantitative analyst. Write a Monte Carlo simulation in Python that models my cash runway. Use the ranges and distributions I specify for revenue growth, costs and collection delays, set a random seed so runs repeat, run [10,000] trials, and report the share of trials that run out of cash before month [NUMBER] and the 10th, 50th and 90th percentile balances. Explain each distribution choice I should confirm. [DESCRIBE YOUR ASSUMPTION RANGES]

SpecializedModel reviewer

Act as a model reviewer. Audit the model below for hard-coded numbers inside formulas, inconsistent formulas across a row, wrong absolute or relative references, circular references, unit mismatches and totals that do not reconcile. List each problem with its cell or line, why it matters, and a fix, ordered by impact on the result. Do not change anything until I approve. [PASTE YOUR FORMULAS OR CODE]

What to add to any financial modeling prompt#

The numbers and their source. Paste historicals and name the period and source. Anthropic notes that Claude lacks context on your norms and workflows, and OpenAI says to place supplied data in the prompt.
The layout you want. Ask for inputs, calculations and outputs on separate sheets with consistent rows. The FAST Standard calls for rigorous consistency in layout and organisation, and for simple, clear formulas.
Spreadsheet or code, and the exact version. Say Excel, Google Sheets or Python, since reference styles differ. In Excel, A1 is relative and $A$1 is absolute, which decides whether a copied formula shifts.
A checks block. Ask for explicit tests such as balance sheet balances and ending cash equals the cash flow statement. Anthropic suggests asking Claude to verify its answer against test criteria.
Wrapped inputs. Put assumptions, data and instructions in separate tagged blocks. Anthropic says tags reduce misinterpretation when a prompt mixes instructions, context and inputs.
An instruction to flag what is missing. Ask the model to list assumptions it had to invent, rather than fill gaps silently.

A worked example#

This fills in the SaaS prompt for a small subscription business.

Code

Act as a startup CFO. Build a monthly SaaS forecast for 24 months from my starting figures: new customers per month, price, monthly churn, expansion, acquisition cost and gross margin. Calculate MRR, ARR, CAC, LTV and payback period, show each formula, and tell me which two assumptions move the result most.

<assumptions>
Starting customers: 120. New customers per month: 25, growing 3% a month.
Price: $49 per customer per month. Monthly churn: 4%. Expansion: 0.
Blended CAC: $310. Gross margin: 78%.
Opening cash: $180,000. Fixed monthly costs: $22,000.
</assumptions>

Put these on an Inputs sheet. Add a checks row: customers at month end = start + new - churned. List any assumption you added.

What a good result contains:

A customer roll-forward that reconciles every month, with churned customers calculated from the 4% input, not typed in.
MRR equal to customers times price, and formulas that reference input cells, not copied numbers.
LTV and payback shown with their formula and gross margin applied, so you can recompute them by hand.
A short statement of which assumptions (here, likely churn and CAC) the result is most sensitive to, which you can confirm by changing them.
A list of anything the model added that you did not supply.

The full prompt library is on our prompt library page.

Limits of this guide#

This page covers building and testing a model from assumptions you provide. It does not cover valuing listed companies from market data, which our stock research prompts address. The FAST Standard page we read states four principles, not detailed rules; we did not read its full PDF. We did not test the prompts or measure whether any model gets the arithmetic right.

Sources & Further Reading#

All sources fetched on 2026-10-22. Pages for OpenAI's guide were read directly, not via the archive, because the archive was not reachable from our tool.

Claim	Source	Section or date
Models need clear, explicit instructions and lack your context	Prompting best practices (Anthropic)	Be clear and direct; undated
XML tags reduce misinterpretation in mixed prompts	Prompting best practices (Anthropic)	Structure prompts with XML tags; undated
Ask the model to verify its answer against test criteria	Prompting best practices (Anthropic)	Self-check; undated
Supplied data goes in the prompt; examples should cover varied inputs; build evals	Prompt engineering (OpenAI)	Context, examples, evals; undated
State clear constraints, context and response format	Prompt design strategies (Google)	Instructions, constraints, context, format; undated
Consistent layout and simple, clear formulas	The FAST Standard	Structured and Transparent principles; undated
Relative `A1` and absolute `$A$1` references behave differently when copied	Overview of formulas in Excel (Microsoft)	Cell references; undated