Every financial model produces a base case. The base case is the row of numbers where every assumption lands roughly where the founder expects. It is also, by itself, close to useless for making a decision, because it answers the wrong question. The question a base case answers is “what happens if I am right about everything.” The question that determines whether you should commit capital is “what happens when I am wrong, and how wrong can I afford to be.” Sensitivity analysis is the only part of the model that answers the second question.
This is why experienced investors skip past the base case and go straight to the sensitivity tables. The base case tells them what you hope. The sensitivity analysis tells them what you understand. A founder who has mapped where their own model breaks is a founder who has thought about the business. A founder who presents only a clean base case has presented a wish.
What sensitivity analysis actually does
Sensitivity analysis takes the inputs that drive the model and varies them, one at a time and then in combination, to see how far the output moves. The point is not to produce more scenarios. The point is to find the breaking point: the value of each key variable at which the business stops working, runs out of cash, breaches a covenant, or fails to clear its return threshold.
A well-built sensitivity analysis tells you three things for each variable that matters. How much the outcome moves for a given change in the input. The exact input value at which the outcome crosses from acceptable to unacceptable. And how much margin sits between your base-case assumption and that crossing point. The third number is the one that matters most, because it is the answer to “how wrong can I afford to be.”
Pick the variables that actually move the outcome
Not every input deserves a sensitivity test. Varying an assumption that barely touches the result wastes pages and buries the variables that matter. The discipline is to identify the small number of inputs that swing the outcome most, usually three or four, and concentrate the analysis there.
For most businesses the high-leverage variables are some combination of price, volume or conversion, a primary cost input, and the timing of revenue. For capital-intensive ventures the interest rate and the ramp speed join that list. The tornado chart exists precisely to rank these. It shows, at a glance, which variables produce the widest swing in the outcome, which is the same as showing which assumptions the reader should challenge hardest. A model without a tornado chart forces the reader to guess where the leverage is. A model with one hands them the map.
Single-variable, then combined
Start with one variable at a time. Hold everything else at base case and walk price, then volume, then the cost input, then timing, across a realistic range. This isolates the effect of each and produces the clean tornado ranking. It is the version most models stop at.
The version that actually tests feasibility goes one step further and combines the downside moves. The real risk is rarely a single variable missing in isolation. It is two or three missing together, because the conditions that soften your price often also soften your volume. A combined downside case, where the two or three highest-leverage variables all land below plan at the same time, is the honest stress test. If the model survives the combined downside with cash to spare, the deal is robust. If it survives single-variable shocks but breaks the moment two land together, the base case has been hiding a fragile structure.
The breakeven is the number to name
For each high-leverage variable, the most useful single output is the breakeven: the value at which the outcome crosses zero, or crosses the covenant, or crosses the return hurdle. “We breakeven at 62% of our assumed conversion rate” is a sentence an investor can reason about immediately. It locates the model against reality. If comparable businesses convert at 70% and you breakeven at 62%, you have a thin but real margin. If they convert at 55% and you breakeven at 62%, the base case is assuming a level the market does not support, and the sensitivity analysis just exposed it.
Naming the breakeven also disarms the most common objection. When a reader challenges an assumption as too optimistic, the breakeven answers before the argument starts: here is how much that assumption can be wrong before the deal stops working. The conversation moves from “I do not believe your number” to “is this margin of safety enough,” which is a far more productive place to be.
Where founders get it wrong
Three failures recur. The first is testing only the upside, varying assumptions in the favorable direction to show how good things could get. That is not sensitivity analysis. It is a sales pitch with extra columns. The second is varying everything by a uniform percentage, plus or minus 10% across all inputs, which ignores that some variables realistically move far more than others and treats a 10% price change and a 10% rate change as equivalent risks when they are not. The third is presenting the sensitivity tables without naming the breaking points, leaving the reader to compute them. The reader will not compute them. They will assume you did not, and they will discount the whole model.
Why this belongs in the feasibility study, not after it
Sensitivity analysis is not a polish step applied to a finished model. It is the reason the model exists. A feasibility study is supposed to find the point of failure before capital funds it, and the sensitivity analysis is the instrument that does the finding. Running it last, as a formality, defeats the purpose. Running it as the test, with the willingness to let it produce a do-not-proceed, is what separates a feasibility study from an illustration.
The Feasibility Study Standard Pondera ships covers market, financial, operational, and risk analysis across 20 to 25 pages, and the financial section includes the sensitivity work described here: the high-leverage variables identified, single-variable and combined downside cases modeled, breakeven points named, and an explicit go or no-go recommendation that the analysis actually supports. The fee is the same whether the recommendation is proceed or do not proceed. Five business days.
A base case shows what you hope will happen. The sensitivity analysis shows what you can survive. The second is the one the decision rests on.