Read a Tornado as a comparison of tested ranges

Choose the shortlist and defend its ranges

Use the annual operating fixture: price EUR 10, volume 100 units, unit COGS EUR 5 and fixed opex EUR 300. EBITDA is EUR 200. Select these four eligible 2027 inputs for Model EBITDA in Base. Price 9–11 and volume 80–120 are not equally sized percentage changes: the first is ±10%, the second ±20%.

A plausible range should come from an explicit planning question, contract boundary or evidence. Equal percentage ranges can help comparison, but do not make different business uncertainties equally credible. State the unit beside each range.

Read direction as well as bar length

Price and volume each span EUR 200 in this test. Unit cost 4.5–5.5 and fixed opex 250–350 each span EUR 100. Higher price or volume raises EBITDA; higher costs reduce it, so a low input can be the high output. The table preserves native ordering for tied spans; ties do not support a first-versus-second importance claim.

The span includes Base and both endpoints: max(low output, Base, high output) − min(low output, Base, high output). It is not necessarily twice the largest deviation from Base, and it does not add together with other bars.

Widen one range and the ranking changes

Change only the price range to EUR 8–12. Native EBITDA endpoints become EUR 0 and EUR 400, giving a EUR 400 span. Price now outranks volume because its test became wider. The company and Base EBITDA did not change.

Keep the original ranges when comparing periods or scenarios, or explain why they changed. Otherwise the apparent change in importance may come entirely from your range selection.

1. Select the target and open Analyze → Tornado.
2. Add the four eligible inputs and enter the explicit low/high ranges.
3. Run, inspect both endpoints and Base, then compare spans.
4. Widen price only, rerun and record the changed experiment.

Keep Tornado distinct from discovery and probability

Key Drivers discovers eligible relationships and ranks the largest absolute tested deviation from Base, including joint group tests. Tornado starts from a shortlist you selected and compares total spans. Neither establishes real-world causality.

Endpoints do not describe the shape between them or the frequency of either endpoint. An unavailable endpoint is a failed test, not a zero impact. This native example verifies all endpoints and restores the exact workspace after calculation.

Related decisions and model reference

For adjacent questions: Financial Model Analysis — Drivers, Margins and Cash; Operating Leverage Analysis: Sales Scenarios to EBIT; Monte Carlo vs Sensitivity Analysis: Probability or Response?.

For the exact controls and source behavior, use Tornado: compare selected ranges and Key Drivers: discover tested impact.