Verified against ChatGPT · 2026-08-12
Stress-test a financial model's assumptions page before someone else finds the weak one first
Runs a structured stress test on the assumptions driving a financial model, ranking them by how much the output actually depends on each one, so the shakiest assumption gets caught before an investor or exec finds it.
The prompt
Ready to copy — highlighted parts are example details you can swap.
I'm about to share a financial model and I want the assumptions page stress-tested first — specifically, I want to know which assumptions the output is most sensitive to, before someone reviewing it finds the weak one and the whole model loses credibility over one bad input.
MODEL PURPOSE
18-month cash flow and headcount forecast for a Series A pitch
KEY ASSUMPTIONS AS LISTED
15% MoM revenue growth; 4% monthly gross churn; hiring 2 engineers per quarter at $160k loaded cost each; 70% gross margin held flat
OUTPUT THE MODEL PRODUCES
Shows the company reaching cash-flow breakeven in month 14 without an additional raise
WHO WILL REVIEW THIS
A Series A investor's associate doing diligence, known for asking pointed unit-economics questions
For each listed assumption, do three things. First, classify it as either a hard input (something observable or contractually fixed, like a signed price or a known headcount) or a judgment call (something projected or estimated, like a growth rate or a churn assumption) — judgment calls are where real scrutiny will land, and I want them clearly distinguished from hard inputs. Second, for each judgment call, estimate roughly how sensitive the model output is to a reasonable swing in that assumption (e.g., what happens to the output if this assumption is off by the kind of margin a projection like this typically is) — rank the judgment calls from most to least sensitive. Third, for the two or three most sensitive judgment calls, propose the specific, pointed question a skeptical reviewer would ask to test it ("where did the 8% monthly growth assumption come from, and does it hold if the biggest customer doesn't renew") — write these as the actual questions to prepare for, not vague advice like "be ready to defend your growth assumption."
WHAT NOT TO DO
Do not invent a specific sensitivity percentage or dollar swing as if you calculated it precisely — you don't have the underlying model mechanics, only the assumptions and stated output, so describe sensitivity directionally and qualitatively ("the output is disproportionately dependent on this one") rather than fabricating a specific number that looks like real modeling output. Do not tell me an assumption is reasonable or unreasonable — that's a judgment call for me and whoever set it, not something you can assert from a list of numbers alone.
OUTPUT FORMAT
1. A table: Assumption | Hard Input or Judgment Call | Sensitivity Rank (if judgment call) | Why.
2. The prepared-questions list for the top 2-3 most sensitive judgment calls.
3. One line noting this is a structuring aid to prepare for review, not a substitute for someone with real financial modeling expertise actually rebuilding the sensitivity analysis inside the model itself.Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
Separating hard inputs from judgment calls before ranking anything is the structural move that makes this stress test useful rather than performative — a reviewer scrutinizing a model is not going to spend time questioning a signed contract price, they're going to go straight for the projected growth rate and churn assumption, and pre-sorting the list into these two categories mirrors exactly where real diligence attention lands, rather than treating every line item as equally worth defending. Explicitly instructing the model to describe sensitivity qualitatively and directionally rather than generating a specific percentage swing is the most important guardrail in this prompt: GPT-5.1, asked for a sensitivity analysis, can produce a fluent, numerically specific-looking answer ("a 3-point swing in churn changes breakeven by 2.4 months") that sounds like it came from actually running the model, when in fact the model was never given the underlying formulas to compute that — presenting a fabricated precise number as if it were real modeling output is more dangerous than an honest qualitative flag, because it would be repeated in a pitch as if it had been verified. Producing the actual pointed questions a skeptical reviewer would ask, rather than generic advice to "be ready to defend your assumptions," is what makes this genuinely useful preparation — a founder walking into diligence needs the literal question stated ("does breakeven still hold if the biggest customer doesn't renew") so they can pre-build the answer, not a reminder that scrutiny is coming. The closing instruction that this doesn't replace someone with real modeling expertise rebuilding true sensitivity analysis inside the model is an honest scope limit: a language model reasoning over a list of stated assumptions cannot actually recompute the model's real mechanical sensitivity, only reason about which inputs are inherently more speculative than others.
What you get back
Assumption: 15% MoM revenue growth | Judgment call | Sensitivity: highest | Why: compounds over 18 months, so a modest miss early compounds into a large deviation in the breakeven month, and it's the single most speculative number in the set relative to the hard-input hiring costs. Prepared question: "What does breakeven look like if growth runs at 10% instead of 15% for the first two quarters, and does the current cash position still cover the gap?"
Verified against
ChatGPT GPT-5.1 · 2026-08-12
Changelog
- 2026-08-12 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
