AI DCF prompt checklist
Market participants are sharing AI prompts to generate full DCF models that include revenue/expense projections, a WACC built from beta and equity‑risk premium, terminal value via Gordon or exit multiples, and sensitivity tables (x.com) (x.com). The recommended output packages bull/base/bear cases and sensitivity matrices so the model reads like a sell‑side pitch book and becomes decision‑ready (x.com) (x.com).
A discounted cash flow model is supposed to be the slow part of stock work: you forecast sales for 5 to 10 years, turn those sales into free cash flow, and discount each year back to today. Finance training materials still teach it as a step-by-step build, usually in Excel, because one broken assumption can swing the answer by billions. (cfainstitute.org) (corporatefinanceinstitute.com) What changed is that investors are now passing around prompts that ask an artificial intelligence model to build the whole thing in one shot: revenue build, margin path, free cash flow, discount rate, terminal value, and a finished write-up. The examples circulating on X explicitly ask for bull, base, and bear cases plus sensitivity tables so the output looks like a sell-side research book instead of a rough worksheet. (x.com 1) (x.com 2) (x.com 3) The reason that prompt format works is simple: a discounted cash flow model is mostly a checklist. You need a forecast period, a discount rate called weighted average cost of capital, and a terminal value that captures the years after your explicit forecast ends. (corporatefinanceinstitute.com 1) (corporatefinanceinstitute.com 2) (corporatefinanceinstitute.com 3) Weighted average cost of capital is just the blended return demanded by the people funding the business. In standard finance teaching, the cost of equity often comes from the Capital Asset Pricing Model, which uses beta as a measure of stock volatility and adds an equity risk premium on top of a risk-free rate. (corporatefinanceinstitute.com) Terminal value is the part that usually dominates the answer. Corporate Finance Institute notes that analysts typically estimate it either with a perpetual growth method, often called the Gordon Growth Model, or with an exit multiple that applies a market valuation multiple at the end of the forecast period. (corporatefinanceinstitute.com 1) (corporatefinanceinstitute.com 2) That is why the new prompts keep asking for sensitivity tables. If you change the weighted average cost of capital by a point or nudge the terminal growth rate, the valuation can move fast, so the matrix is the part that shows whether the thesis survives small mistakes. (corporatefinanceinstitute.com) (corporatefinanceinstitute.com) The software side of this story is not just “ask a chatbot for a number.” OpenAI’s current documentation pushes developers toward explicit prompt structure and Structured Outputs, which return machine-readable fields in a fixed schema instead of a loose paragraph that can skip a line item or rename a key. (openai.com) (developers.openai.com) That matters because a valuation model is really a stack of assumptions pretending to be arithmetic. If the model returns revenue growth, operating margin, tax rate, net working capital, capital expenditures, weighted average cost of capital, and terminal growth in a clean schema, an analyst can audit each field instead of trusting one polished narrative. (developers.openai.com) (developers.openai.com) So the real shift is not that artificial intelligence discovered a new way to value companies. The shift is that the first draft of a full discounted cash flow package, including scenario cases and presentation-ready tables, can now be generated in minutes, which moves the bottleneck from spreadsheet construction to assumption checking. (x.com) (x.com) (cfainstitute.org) That also explains the catch. A language model can format a beautiful valuation book, but it cannot know whether a chip company deserves a higher beta than a software company, whether a 4 percent perpetual growth rate is absurd, or whether management guidance from last quarter is already stale unless a human or a connected data pipeline checks those inputs. (corporatefinanceinstitute.com) (corporatefinanceinstitute.com) (developers.openai.com) The people sharing these prompts are really standardizing a workflow: tell the model exactly which assumptions to surface, force it to show bull, base, and bear outcomes, and make every fragile number visible. That turns artificial intelligence from a magic-answer machine into a very fast junior analyst with a checklist and a formatting obsession. (x.com) (openai.com)
