What it calculates

Sensitivity analysis measures how much an output changes when an input changes, one variable at a time. It identifies which assumptions actually drive your answer and which barely matter, so scrutiny goes where it counts.

How it is calculated

Hold every input at its base case, vary one across a plausible range, and record the output at each step. Repeat for each input. Ranking the resulting output swings produces a tornado chart, widest bar at the top. A two-way table varies two inputs simultaneously across a grid, which is how the classic DCF valuation table of discount rate against terminal growth is built.

How to read the result

The ranking matters more than any individual number: if the top two inputs explain most of the variation, that is where research and negotiation should go, and the rest can be set to reasonable defaults. Be careful that the ranges are equally plausible - a variable looks influential simply because you swung it further. And remember that one-at-a-time analysis misses interactions, which is exactly what scenario and Monte Carlo analysis are for.

Worked example

A base NPV of 12,000,000 moves to between 6,400,000 and 18,900,000 when the discount rate varies by plus or minus two points, but only to between 10,800,000 and 13,300,000 when working capital days vary by plus or minus ten. The discount rate is roughly five times as influential, so the cost of capital deserves the attention.

Common questions

What is the difference between sensitivity and scenario analysis?

Sensitivity varies one input at a time, which isolates each driver but describes a combination that may not be coherent. Scenario analysis varies a set of inputs together in an internally consistent story - a recession has low growth and high discount rates and weak collections at once.

How wide should the ranges be?

Wide enough to cover realistic uncertainty and no wider. A useful discipline is to use a genuine confidence interval for each input rather than a uniform plus or minus ten percent, which arbitrarily makes stable and volatile inputs look equally uncertain.

Why is it called a tornado chart?

Because the bars are sorted from widest to narrowest around a central base-case line, giving a funnel shape. The sorting is the point: it puts the assumptions that matter at the top where they are read first.