Capital budgeting remains one of the most consequential and most tedious responsibilities in corporate finance. Every year, controllers and financial analysts receive dozens of project proposals — new equipment, facility expansions, software migrations, acquisitions — each with its own assumptions, timelines, and cash flow estimates. Building a defensible NPV, IRR, and payback model for every single proposal can consume weeks of analyst time, and the output is often a patchwork of inconsistent spreadsheet conventions, hardcoded assumptions, and copy-paste errors that nobody catches until an audit or a board meeting.
The friction is not the math. NPV and IRR are well-understood formulas. The pain lies in the surrounding work: normalizing input data from different departments, documenting assumptions consistently, stress-testing discount rates, comparing projects on equal footing, and producing a narrative that a CFO or investment committee can actually act on. Analysts spend 80 percent of their time on formatting, reconciliation, and rework, and only 20 percent on the analysis that changes the decision.
AI tools — particularly large language models like Claude — change this equation. When given the right context files and a structured prompt, an AI assistant can ingest raw project data, apply your firm’s discount rate policy, generate a clean NPV/IRR/payback table, run sensitivity scenarios, and draft the investment memo — all in a fraction of the time. The analyst shifts from builder to reviewer, and the quality of the output actually improves because the logic is documented and reproducible.
Why a structured prompt matters here
Capital budgeting is a domain where generic AI prompts fail. If you simply ask an AI to “calculate NPV for this project,” you will get a plausible-looking answer built on assumptions the model invented. In finance, invented assumptions are worse than no answer. The fix is to supply your own standards — discount rate policy, hurdle rates, tax treatment, depreciation conventions, terminal value rules — as context files, and to force the AI to ask clarifying questions before it starts computing.
The two prompts below are built around that principle. The first handles the core analytical engine: turning raw project inputs into a clean, auditable NPV/IRR/payback model. The second handles the communication layer: converting that model into an investment committee memo that a CFO can sign off on.
First, read these files completely before responding:
[capital_policy.md] — our firm’s discount rate policy, hurdle rates by project category, tax treatment, and depreciation conventions
[project_inputs.md] — raw cash flow estimates, capex schedule, opex assumptions, and revenue projections for this project
[model_template.md] — the exact table structure and column headers our committee expects to see
Here is a reference for what I want to achieve:
[Upload a prior approved capital budgeting memo as markdown, or describe the structure of your last approved model]
Here’s what makes this reference work:
It leads with a one-page summary table (NPV, IRR, payback, profitability index), then shows the full cash flow schedule year by year, then a sensitivity grid on discount rate and revenue growth, then a short risk section. Assumptions are footnoted, not buried. Every number traces back to a named input.
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: Structured markdown analysis, roughly 800-1200 words plus tables
Recipient’s reaction: “This is audit-ready, I can forward it to the board without edits”
Does NOT sound like: A generic finance textbook, a wall of formulas without interpretation, or a model with unexplained assumptions
Success means: The committee can make a go/no-go decision in a single meeting using only this document
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.
Notice what that prompt does. It forces the AI to read your policy file before touching the numbers, it anchors the output to a reference document the committee already trusts, and it defines success as a decision, not a calculation. The clarifying-questions step is critical — it surfaces missing inputs (residual value, working capital release, tax loss carryforwards) before the model is built, not after.
From model to memo: the second half of the job
Even a perfect NPV model fails if the memo around it is unreadable. Investment committees do not want to reverse-engineer your spreadsheet; they want the thesis, the numbers, the risks, and the recommendation in that order. The second prompt below handles that translation, and it is designed to work directly off the output of the first prompt.
First, read these files completely before responding:
[committee_memo_template.md] — the standard structure, section order, and tone our investment committee expects
[prior_approved_memo.md] — a memo that was approved last quarter, as a style reference
[risk_register.md] — our standard categories of project risk (execution, market, regulatory, financing)
Here is a reference for what I want to achieve:
[Upload the approved memo you want to emulate, or describe its structure]
Here’s what makes this reference work:
It opens with a three-sentence recommendation, then a summary table, then a “why now” paragraph, then a risks-and-mitigants section, then a clear ask. No jargon without definition. No number appears without context.
Here’s what I need for my version / SUCCESS BRIEF:
Type of output + length: Executive memo, two pages maximum, roughly 600-800 words
Recipient’s reaction: “I understand the tradeoff and I know what I’m voting on”
Does NOT sound like: An academic paper, a sales pitch, or a data dump
Success means: The committee approves, defers, or rejects with a documented rationale in one meeting
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.
The practical tip here is to run both prompts in the same conversation. The second prompt inherits the full context of the first, so the memo will reference the exact NPV, IRR, and sensitivity figures the model produced — no manual transcription, no risk of the memo and the model disagreeing. If your AI tool supports project files or memory, keep your capital policy and memo template loaded permanently so you only update the project-specific inputs each time.
What to try next: pick one live project currently in your pipeline and run it through both prompts end to end. Compare the output side by side with whatever your team produced manually. In most cases the AI version will be faster, more consistent, and — because the assumptions are explicit — easier to defend. From there, standardize the two prompts as your default capital budgeting workflow and let your analysts spend their time on the judgment calls that actually require a human.
Published on 11 September 2026 on growwithgpt.com