Tornado Chart

A business case depends on customer growth, selling price and construction cost, and managers want to know which assumption most affects the forecast. A tornado chart shows the range of outcomes when each one changes in turn.

Oracle's tornado analysis tests input ranges and plots the swing in forecast values, sorted from largest to smallest, and that ordering gives the chart its shape. Start by choosing the outcome, because profit, cash flow, NPV (net present value) and completion cost tell different stories, and label the chart and units clearly.

Then build a base case that records the model's starting assumptions and result, since a precise chart will still mislead if the model is unreliable. Choose inputs that decision-makers can influence together with important uncertainties they cannot, and avoid adding many duplicates of the same driver.

Set plausible low and high values with a documented basis, because an arbitrary wide range can make one factor appear dominant. Change one input at a time while holding the others at base values, which isolates a local sensitivity, and run the model at each endpoint.

The difference between the two results is the displayed swing under a simple method, and the bars are sorted with the larger swing near the top so the outline resembles a tornado. Show direction and baselines: a higher price may raise profit while a higher cost may lower it, so colour and labels should make this clear without relying on colour alone, and the base-case result should be marked so viewers see how far each case moves it.

Check for nonlinear behaviour, since a model can have thresholds or curved responses and testing only two endpoints may miss a peak or trough in between. Compare unit scales as well, because a one-percent price change and a ten-percent cost change are not equal shocks, so explain the ranges to prevent false ranking.

A wide bar means a large possible effect across the chosen range, not that the outcome is likely, and one-at-a-time changes do not capture correlations such as price and volume moving together. Use scenarios too, since a combined downside case can reveal what happens when several inputs worsen at once, and do not add isolated swings mechanically.

A tornado chart shows sensitivity under tested values, whereas a full risk analysis may include likelihood and dependencies. Avoid false precision, because a 1.37 million swing can look exact when the underlying sales forecast is rough, so round to an appropriate level and give market prices, project estimates and operational forecasts dates and owners.

A surprisingly wide bar can reveal a spreadsheet formula error, so validate the model before presenting, and refresh the chart as estimates improve or the decision changes. For owners, the chart directs attention to the assumptions that matter most and shows where tighter management or hedging may pay off, but it is a ranking tool, not a forecast, so keep the tested ranges visible beside each bar.