sensitivity analysis

Also called: what-if analysis

Changing one assumption at a time to see how much the result moves, to find the assumptions a decision really depends on.

Sensitivity analysis changes one assumption at a time, keeps the others at the base case, and measures how far the result moves. Using the same step for every driver (say 10%) makes them comparable; sorting them by impact (often drawn as a tornado chart) shows which assumptions the decision really depends on.

The ranking is often surprising. In a business with a thin margin, price and unit costs dominate, because 10% of a big flow beats 10% of a small leftover; volume matters less than the planning meeting spent on it. Step costs can make a driver's effect uneven, so it pays to recompute through the model rather than multiply.

A second form turns the question around: how far can one driver move before the result hits zero? That break-even value is a red line the team can watch. Sensitivity has a limit: in reality drivers move together (a weak market brings fewer customers and lower prices at once), which is what scenario analysis is for.

Example

Klasna's forecast for next year gives 2,251,968 UAH of operating profit. Each driver moved by 10% in the direction that matters, the rest at the base case:

Driver moved by 10%	Change in profit, UAH
Price per lesson +10%	+5,078,112
Tutor payout +10%	−3,325,312
Lessons per learner +10%	+1,500,434.25
Active learners +10%	+1,424,434.25
Marketing −10%	+309,000
USD rate +10%	−252,403.20

Price: 80,128 lessons × 65 × 0.975 = 5,078,112. Learners add less than lessons per learner because two months cross a coordinator threshold (+76,000). The break-even average price is 621.17 UAH, only 4.43% under the planned 650: the one red line to watch.

Common mistakes

Using different steps for different drivers. Compare drivers at the same step, or at realistic ranges stated for each.
Multiplying instead of recomputing. Step costs and thresholds make effects uneven.
Stopping at the table. Turn the top driver into a break-even value and a signal to watch.
Treating one-at-a-time as the whole risk. Drivers move together in real years; follow up with scenarios.
Testing only the drivers someone argues about. The noisiest assumption is rarely the one that matters most.

Learn it in the course

Scenarios and sensitivity · Which assumption matters most, and a plan for the weak case.
Case: build our own video classroom or keep paying per minute? · A make or buy decision in cash: two options, their incremental cash flows, the NPV and the assumptions that decide it.