What-if analysis: turn one dial at a time
What-if analysis means changing one or a few inputs of a model and seeing how the outputs move. Most people first meet it in a spreadsheet. Microsoft's guide to Excel describes it as changing the values in cells to see how the formulas respond, and lists three built-in tools:
- Scenarios save sets of input values that you can swap in and compare.
- Goal Seek works backwards: you set the result you want and it finds the input that gets you there.
- Data Tables show the result for many values of one or two inputs at once.
The model stays the same. Only the inputs change. That makes what-if analysis fast, precise and easy to explain. Its weak spot is that it assumes the structure of the world stays as it is. It tells you what happens if fuel costs rise by a tenth. It does not tell you what happens if the whole way people travel changes.
Scenario analysis: imagine several whole futures
Scenario analysis works at a different level. A scenario is a coherent story of a possible future, where many things change together and in a consistent way. The UK Government Office for Science, in its Futures Toolkit for policymakers, is blunt: "Scenarios are not predictions". No single scenario will come true, but parts of several probably will.
The toolkit also gives practical advice. Scenarios should look at least five years ahead, and usually 10 to 20. A common way to build them is a 2×2 matrix: pick the two most important and most uncertain drivers, use them as axes, and describe the four futures in the corners. The scenarios can then be used to stress-test a policy: does the plan still work in each of the four worlds?
Side by side
| What-if analysis | Scenario analysis | |
|---|---|---|
| Question | What happens to the result if this input changes? | Which futures could we face, and does our plan survive them? |
| What changes | One or a few numbers | Many linked drivers at once |
| Time horizon | Often months to a few years | Usually 10 to 20 years |
| Output | Numbers, sensitivities, break-even points | A small set of stories, often with numbers attached |
| Main risk | Missing a change in the system itself | Treating one scenario as the forecast |
Example: a municipality planning for summer
Imagine an island municipality that worries about water in August. The figures below are made up for illustration.
What-if: "If visitors in August rise by 15%, how much does peak daily water demand rise, and on how many days would we exceed what the network can deliver?" This is a clear question for a model of today's network and demand.
Scenarios: "What could summer look like in 2040?" Take two uncertain drivers, such as how dry the climate becomes and how fast tourism grows. The four corners might be: dry and crowded, dry and quiet, wet and crowded, wet and quiet. For each, the municipality asks whether its plans hold up: a new storage tank, leak repairs, a desalination unit, or water pricing. A good plan is one that is acceptable in all four, not perfect in one. We apply this thinking to Greece's water in a separate post on water shortage in Greece.
Example: a business
What-if: a hotel group asks how its margin changes if energy prices rise by 20% next season. A data table answers this in minutes.
Scenarios: the same group asks what its market could look like in 15 years, with axes such as the number of heatwave days in its region and whether travellers shift to shoulder seasons. The answer is not a number but a set of futures to check investments against, such as cooling, insurance or where to build next.
Where simulation comes in
Both methods need a model that can be run. In a spreadsheet that model is a set of formulas. For a city, a coast or a power grid it has to be richer, and this is where simulation and digital twins come in. A twin can run a what-if inside each scenario: "in the dry and crowded future, what if the main pipeline fails for a day?" That joins the precision of what-if analysis with the breadth of scenarios. We describe the general idea on our how it works page.
Three common mistakes
- Calling a what-if a scenario. Moving one number up and down is sensitivity testing. It does not explore different futures.
- Planning for the middle scenario. The point of scenarios is to prepare for the range. A plan built only for the "central" case is a forecast in disguise.
- Never checking back. Revisit scenarios as the years pass. Which one is the world drifting towards? That is useful information too.