How it is calculated
Visitors enter the funnel and each step passes a share of them to the next one. Whoever reaches the end spends the average order value. That is the whole model:
Orders per period equal visitors times every step rate multiplied together. Revenue equals orders times average order value. Annual revenue equals that times the number of periods in a year.
The number that ranks the funnel is the one under each step. Adding one percentage point to a step adds one hundredth of the traffic arriving at it, and every step after it still applies. So a point is worth the traffic reaching that step, divided by 100, times every rate downstream, times the average order value, times the periods in a year.
That is why the biggest number is not always at the top. A point at the top carries more traffic but loses most of it to the steps below. A point at the bottom carries less traffic and keeps all of it. Which one wins depends on your rates, which is the question this answers and a spreadsheet usually gets wrong.
What the range is, and what it is not
The low and high cases are the same arithmetic at half the lift and at one and a half times it. They are a pessimistic and an optimistic case for a pitch. They are not a confidence interval, they carry no probability, and nobody should present them as though a statistical test produced them. If you want a real interval, run the test and read it from the result.
What it does not do
It assumes the steps are independent and that improving one does not move another. In a real funnel they interact: a change that pushes more people into the cart can lower the rate at checkout because the extra people are less committed. It assumes average order value holds steady, which it does not if the change attracts a different kind of buyer. And it models a steady state: the ramp while a change rolls out is nowhere in it.
Use it to decide what is worth testing. Use the test to decide what is true.