Skip to content
Predict.ai

In production

What-if analysis: simulate the decision before you make it

A what-if is a second forecast under a change you control — price up 8%, a promo pulled, a plant offline. You compare paths, not opinions. The point is not a perfect simulation; it is catching the decision that only looks cheap this week.

Updated Aug 4, 2026·7 min read

A second path

A what-if is a second forecast under a change you control. Price up 8%. Promo pulled. Plant offline. You compare paths from the same model, not two opinions in a Google Sheet that cannot hold an interval.

Base path vs promo-on

What-if scenario versus base forecastdecisionpromo onbase
Same model, same horizon. Lime is the scenario ('run the 10% discount from Friday'). The gap is the estimated lift — still a band in production, drawn here as two center lines so you can see the fork.

Levers that are actually yours

The lever must be an input the model has. “What if demand were higher” is a wish. “What if we cut price 8% and keep the learned price response” is a scenario. Capacity, assortment, marketing spend — same test: can you timestamp it, and did the model see that kind of move before?

If the model never saw a 30% discount, do not treat the path as a lab result. Treat it as an extrapolation, and say so.

How to compare without cheating

Hold everything else still. Change one lever, or a package you would actually ship together. Keep the interval — a point-versus-point comparison pretends you know more than you do. Look at the same horizon you would buy on, not a prettier one.

What a scenario cannot do

It cannot see a competitor’s surprise. It cannot invent a relationship that was not in the training window. It cannot replace a test market. What it can do is catch the decision that only looks cheap this week — the promo that steals next week, the price cut that fills the truck and empties the margin.

FAQ

What is what-if analysis in forecasting?
You take the live model and change an input you actually control (price, promo, capacity), then generate a new path. The difference is the estimated effect, with the same uncertainty style as the base forecast.
Is this the same as a sensitivity analysis?
Sensitivity usually nudges one input to see how jumpy the output is. A what-if is a decision-shaped change ('cancel the holiday promo'). You want both: one for trust, one for planning.
Why not just build another spreadsheet?
You can. You will lose the model's learned relationships and the interval. Spreadsheets are great at arithmetic. They are weak at 'what usually happens after a 10% discount on this category.'

Keep going

Try it on your data.

Connect an outcome and its history. Predict.ai finds the drivers, races the models, and keeps the forecast live — in the workspace, over the API, and through agents.