First, the key terms.

Trial
One sampled set of inputs and its resulting outputs.
Dependence
A relationship between uncertain inputs that affects their joint behavior.
Tornado chart
An ordered display of sensitivity effects under a defined analysis method.

The idea, explained

A simulation makes an uncertainty model explicit; it does not discover the correct distribution automatically. Define input distributions, dependence, formulas, decision variables and outputs. Repeated trials approximate the outcomes of that model. A larger trial count reduces sampling noise but does not repair biased inputs or missing mechanisms. Tornado charts often rank one-at-a-time input effects, while simulation-based sensitivity may use other methods. Interpret the chart’s construction before treating its ranking as causal importance.

STEP 1

Represent uncertainty plausibly

Use data or justified elicitation to choose distributions and bounds. Demand, delay and cost can be linked. Sampling them independently when adverse events move them together can understate severe outcomes. Correlation alone may not describe every important tail relationship.

STEP 2

Calculate repeated outcomes

In Excel or Analytic Solver, keep inputs and formulas auditable, set decision values, run trials and inspect means, percentiles and threshold probabilities. Check reproducibility settings where applicable and compare simple limiting cases with hand calculations.

STEP 3

Interpret and challenge results

Report model assumptions, trial count, uncertainty and consequential tail outcomes. Vary distribution and dependence assumptions. A tornado ranking based on a chosen input range changes if that range changes; it is not proof that intervening on the variable will have the predicted causal effect.

Follow a worked example.

A toy demand model assigns 0, 10 and 20 units with probabilities 0.2, 0.5 and 0.3. A stock decision of 10 creates a shortage exactly in the 20-unit scenario under this model. Simulation should approximate a shortage frequency of 30 percent over many trials.

Walk it through, one step at a time

  1. Map a uniform random draw into the three outcome ranges, or use the tool’s discrete distribution capability.
  2. For each trial calculate max(demand − stock, 0). Summarize average shortage and the fraction of trials with any shortage.
  3. Check that results approach the hand-calculated model values, then test whether different probabilities or correlated supply delays change the decision. Do not confuse this check with validation of real demand.

Where people get stuck

More simulation trials can make a wrong model’s answer look impressively stable.

Now try it yourself.

Pause before opening each answer. Explain your reasoning in your own words.

1. Check the core idea

Would doubling the trial count correct an omitted correlation between supply delays and demand?

Reveal the answer & reasoning

No. It only samples the specified model more precisely. The dependence structure must be modeled appropriately.

2. Apply what you learned

What should be checked before interpreting the longest tornado bar as the most important real-world cause?

Reveal the answer & reasoning

Check the output, input ranges, other inputs held fixed and construction method. Sensitivity under a model is not automatically causal evidence.