What is sensitivity analysis?

Sensitivity analysis asks how much the answer moves when an input moves. It shows which assumptions matter and which do not. This tells a team where to spend effort on better data and which numbers are safe to estimate roughly.

How it works

The simplest form changes one input at a time by the same percentage, such as plus and minus 10 percent, and records the change in the output. Inputs that cause big swings are sensitive, and inputs that cause small swings are not. A more thorough approach varies several inputs together, because interactions can make the total effect larger or smaller than the sum. Results are often shown as a ranked bar chart. An input being highly sensitive does not mean that it is likely to change, so combine sensitivity with how uncertain each input really is.

Once the sensitive inputs are known, you can spend your effort where it counts, by collecting better data on them or by designing the plan so that it depends less on them.

Example

Say a subscription model gives yearly profit as customers times price minus costs. Customers are 10,000, price is 100 a year, and costs are 700,000, so profit is 300,000. A 10 percent rise in price gives 1,100,000 minus 700,000, which is 400,000, a 33 percent gain. A 10 percent rise in customers gives the same 400,000. A 10 percent rise in costs reduces profit by 70,000, a 23 percent loss. Price and customers matter most. This is an illustration with round numbers.

How to model it in Machinations

Expose each input as a Lever and keep the output in a Register or Pool. Play the diagram once at the base values, then change one Lever at a time by the same percentage and note the output. Rank the changes to see which Levers matter most.