Introduced on day 1
Causal effect
The difference in an outcome between acting and not acting on the same units, for example how many more seats sell if fares drop 5%. It needs an experiment or explicit causal assumptions; a predictive model alone does not supply it.
In detail
A causal effect compares two worlds that never both exist: the flight priced at EUR 120 and the same flight priced at EUR 114. Because one of them is always unobserved, the effect has to be estimated through randomization (experiments) or through assumptions about how the data were generated (Day 5).
Historical airline data are a classic trap. Fares are set higher when demand is expected to be high, so in the raw data high fares go with full flights. A model trained on that history can show demand rising with price, the wrong sign for a pricing decision. The pattern is real for prediction and misleading as an effect.
Ask for a causal effect whenever the question contains an action: "what if we overbook two more seats?", "what if we send this offer?". Then name how you will identify it before you choose a model.