Every financial analyst knows the drill. You open a spreadsheet, pull the latest cash flow projections, and begin the manual grind of computing Net Present Value (NPV), Internal Rate of Return (IRR), and Payback Period. You double-check discount rate assumptions, verify that the IRR formula isn’t returning a #NUM! error, and manually adjust for uneven cash flows. Then you do it again for the second scenario. And the third. By the time you present your capital budgeting recommendation to the CFO, you’ve spent six hours on mechanical computation—and maybe thirty minutes on actual strategic thinking.
The friction is real. Multi-million-dollar capital allocation decisions hinge on calculations that are tedious, error-prone, and slow. A single misplaced decimal in a cash flow row can flip a “reject” into an “accept” for a $10 million project. Worse, the manual approach makes it nearly impossible to run sensitivity analysis across dozens of scenarios in a timely manner. This is where AI transforms the process. By automating the grunt work of NPV, IRR, and Payback analysis, AI tools allow finance professionals to focus on what matters: evaluating risk, challenging assumptions, and making faster, better capital allocation decisions.
In this post, I’ll show you two ready-to-use prompt templates that turn Claude into a capital budgeting automation engine. You’ll learn how to structure your prompts so the AI ingests your cash flow data, applies your discount rate, computes the three core metrics, and even runs sensitivity analysis—all without you touching a spreadsheet cell.
Why This Matters for CFOs and Controllers
Capital budgeting is the lifeblood of corporate growth. Whether you’re evaluating a new manufacturing line, an IT infrastructure upgrade, or an acquisition, the numbers must be right. Yet a 2024 survey by the Association for Financial Professionals found that nearly 40% of organizations still rely on manual spreadsheet-based capital budgeting, with error rates averaging 2-3% per model. For a $50 million project, that’s a potential $1.5 million mistake. AI doesn’t replace your judgment—it eliminates the mechanical errors so your judgment can operate on clean, accurate data.
Prompt 1: Full Capital Budgeting Analysis with Scenario Comparison
First, read these files completely before responding:
[project_cashflows.csv] — Contains annual net cash flow projections for 5 scenarios (Base Case, Optimistic, Pessimistic, High Discount Rate, Accelerated Payback). Columns: Year, Scenario, Net_Cash_Flow, Initial_Investment.
[discount_rate_assumptions.md] — Lists the weighted average cost of capital (WACC) for each scenario, plus terminal value assumptions where applicable.
[capital_budgeting_policy.md] — Company guidelines: minimum IRR threshold of 12%, maximum payback period of 5 years, and a capital allocation cap of $20 million per project.
Here is a reference for what I want to achieve:
I have attached a reference PDF of a capital budgeting committee report from Q1 2026. It shows NPV, IRR, Payback, and Profitability Index for 4 projects, with a ranking table and recommendation.
Here’s what makes this reference work:
– The report uses a consistent table format: Project Name, Initial Investment, NPV (at WACC), IRR, Payback (years), Profitability Index, and Decision (Accept/Reject).
– Scenarios are compared side-by-side with color-coded cells for Accept (green) and Reject (red).
– The narrative summary highlights the top 2 projects and explains why the third was rejected despite a high IRR (breached payback policy).
– Sensitivity analysis is shown as a small table: “If WACC increases by 2%, NPV changes by X%.”
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: A structured capital budgeting report in markdown format, approximately 800-1000 words, with tables and a final recommendation.
Recipient’s reaction: The CFO should be able to read the report in 3 minutes and immediately understand which projects to approve and why. No ambiguity.
Does NOT sound like: Generic AI output with vague advice. Must cite specific numbers from the data files. No emojis or casual language.
Success means: The report automatically flags any scenario where IRR is below 12% or payback exceeds 5 years, and ranks all scenarios by NPV.
My context file contains my standards, constraints, audience. Read it fully before starting.
DO NOT start executing yet. Ask clarifying questions first.
Give me your execution plan (5 steps max) before you begin.
Why This Prompt Structure Works for Financial Analysis
The anatomy above is deliberately over-engineered for a reason. Finance professionals need precision, not guesswork. By specifying the exact files to read (cash flows, discount rates, policy constraints), you eliminate the AI’s tendency to hallucinate numbers. The reference report ensures the output format aligns with what your committee expects. And the success brief forces the AI to think like a financial analyst—flagging policy violations, ranking by NPV, and delivering a recommendation that a CFO can act on immediately. The “ask clarifying questions” step is critical: it lets the AI confirm assumptions about compounding frequency, tax effects, or terminal value treatment before generating the output.
Prompt 2: Sensitivity and Monte Carlo Simulation for Capital Budgeting
First, read these files completely before responding:
[base_case_cashflows.csv] — 10-year annual net cash flow projections for Project Alpha. Initial investment: $15 million. Columns: Year, Base_Cash_Flow, Revenue_Uncertainty (standard deviation as % of cash flow), Cost_Uncertainty (standard deviation as % of cash flow).
[discount_rate_distribution.md] — Describes the WACC as a normal distribution with mean 9.5% and standard deviation 1.2%, based on the company’s debt/equity mix and credit rating.
[risk_tolerance_policy.md] — States that the company requires at least 75% probability of NPV > 0 for any project above $10 million. Also defines “risk-adjusted payback” as the year by which cumulative probability of positive cash flow reaches 90%.
Here is a reference for what I want to achieve:
I have a PDF of a McKinsey-style risk analysis report that shows a tornado chart (sensitivity of NPV to each input variable), a probability distribution of NPV outcomes, and a table showing the 10th, 50th, and 90th percentile NPV values.
Here’s what makes this reference work:
– The tornado chart ranks variables by impact: Revenue growth rate is the highest sensitivity, followed by discount rate, then operating costs.
– The probability distribution is a histogram with 20 bins, showing the range from worst-case (-$2M) to best-case (+$12M).
– The report includes a clear “Go/No-Go” recommendation based on the 75% probability threshold.
– Risk mitigation actions are suggested for the top 3 sensitivity drivers.
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: A risk-adjusted capital budgeting report in markdown, approximately 700-900 words. Include a sensitivity table (not a chart, since text-only), a probability table showing 10th/50th/90th percentile NPV, and a clear recommendation.
Recipient’s reaction: The CFO should immediately see the risk profile and understand whether the project meets the 75% probability threshold. The output must be boardroom-ready.
Does NOT sound like: Overly technical jargon without practical implications. Avoid phrases like “the model suggests” without quantifying. No emojis.
Success means: The report computes the exact probability that NPV is positive, identifies the top 3 sensitivity drivers, and recommends specific risk mitigation actions (e.g., hedge raw material costs, lock in debt financing rate).
My context file contains my standards, constraints, audience. Read it fully before starting.
DO NOT start executing yet. Ask clarifying questions first.
Give me your execution plan (5 steps max) before you begin.
Putting AI Capital Budgeting Into Practice
These two prompts cover the vast majority of capital budgeting work that finance teams face. The first prompt handles the standard project evaluation and comparison that you’d present at a capital committee meeting. The second prompt goes deeper into risk quantification—something that most spreadsheet models handle poorly or not at all. Together, they transform Claude from a general-purpose chatbot into a dedicated financial analysis tool that respects your company’s policies, discount rates, and risk thresholds.
One practical tip: always include your company’s specific policy constraints in the context files. The AI cannot know that your board rejects any project with a payback period beyond 4 years unless you tell it. Similarly, if your firm uses a hurdle rate that varies by division (e.g., 10% for stable operations, 15% for R&D), include that in your discount rate assumptions file. The more precise your inputs, the more reliable the output. Start with a single project, validate the AI’s calculations against your spreadsheet, then scale to full portfolio analysis.
Try running the first prompt with a project you’re currently evaluating. Compare the AI’s output to your manual spreadsheet. I predict you’ll find the numbers match within rounding error—and the AI will have produced the report in the time it takes you to refill your coffee. Next, explore the sensitivity prompt to stress-test your assumptions. You might discover that a project you thought was a slam dunk has only a 55% probability of positive NPV once you account for revenue uncertainty. That insight alone is worth the price of admission.
Published on July 22, 2026 on growwithgpt.com