DCF Stress Test With ChatGPT
The prompt analysts paste to see which single assumption is doing all the work in a valuation — ranked by sensitivity, with bear/base/bull fair values. Model-agnostic (Claude, ChatGPT, Copilot, Gemini).
For: Analysts · IB · Corp-dev · PETo stress-test a DCF with ChatGPT, paste your assumptions and ask the model to rank them by valuation sensitivity rather than to build the model. A working prompt instructs it to: (1) list every assumption and rank them by how much fair value moves per ±1-unit change; (2) flag the three assumptions where you’re most likely being optimistic versus history and peers; (3) produce bear, base, and bull revenue-growth and margin sets with the resulting fair value each; and (4) name the one thing a skeptical investment-committee member would attack first and how you’d defend it. The single most important line is “do not invent numbers I didn’t give you — if you need a figure, ask,” which stops hallucinated inputs. Sanitize live client or deal names to a codename before pasting, and treat the output as a challenge to your model, not a replacement for it.
Your DCF looks clean. The problem is you can’t see which assumption is quietly carrying the whole valuation — and that’s exactly the assumption a skeptical IC member will find first. AI is a fast, tireless devil’s advocate for a model: it will rank your sensitivities, catch where you’re leaning optimistic, and build scenario sets in seconds. The trick is forbidding it from inventing inputs. Here is the workflow.
The workflow, step by step
- Export your assumptions cleanlyCopy the assumption block — revenue growth, margins, WACC, terminal growth, and the resulting fair value. The model reasons over what you give it; messy inputs create messy outputs.
- Sanitize any live dealReplace real client / target names with a codename before pasting. This is the compliance step for anything active.
- Run the stress-test promptIt ranks assumptions by sensitivity, flags optimism vs history/peers, and returns bear/base/bull fair values — with the hard rule not to fabricate inputs.
- Interrogate the load-bearing assumptionOnce you know which input moves the valuation most, ask: “What would have to be true for this to hold, and what evidence would confirm or break it?”
- Rebuild the number in your modelTake the scenarios back into your own spreadsheet. The AI framed the risk; you own the math and the recommendation.
The exact prompt (copy-paste)
Model-agnostic. Swap the [BRACKETS] and paste your source material into the same chat.
Here is my DCF for [COMPANY] (assumptions pasted below). 1. List every assumption and rank them by how much the valuation moves per +/-1 unit change (sensitivity). 2. Flag the 3 assumptions where I'm most likely being optimistic vs history / peers. 3. Give a bear, base, and bull revenue-growth & margin set with the resulting fair value each. 4. Name one thing a skeptical IC member would attack first, and how I'd defend it. Show your reasoning. Do not invent numbers I didn't give you — if you need a figure, ask.
Get all 7 workflows in one pack
This is one of seven paste-ready workflows in The AI Finance Workflows Pack — the exact prompts, pro moves, and compliance guardrails finance pros run at their desk. Free.
Instant access on the next page + the free weekly brief. Unsubscribe anytime.
Prefer to browse first? Open the Workflows Pack → · Finance Prompt Library →
Frequently asked
Can ChatGPT do a DCF sensitivity analysis?
Yes — if you paste your own assumptions and fair value and ask it to rank each input by how much the valuation moves per unit change, then produce bear/base/bull scenarios. It reasons over the numbers you supply; it should not source or invent figures. Instruct it explicitly not to fabricate inputs, and rebuild the final numbers in your own model. Educational, not investment advice.
How do I stress-test a financial model with AI?
Paste the assumption block and ask the model to (1) rank assumptions by sensitivity, (2) flag where you’re optimistic versus history and peers, (3) build bear/base/bull scenarios with resulting fair values, and (4) name the assumption a skeptic would attack first. The output is a challenge to your model, not a substitute for your own diligence. Do your own research (DYOR).
Is it safe to put a DCF into ChatGPT?
Anonymized, public-company models are generally fine. For live deals, replace client and target names with a codename and strip confidential terms before pasting. Never enter MNPI or client PII into consumer AI tools, and follow your firm’s policy.