Price and promo lift — without a fantasy elasticity

Price and promo are levers

Lift is what happens to the path when you change a price or a promo you actually control. Hotel ADR, a kWh tariff, a fare class, a retail discount: same math family, different politics. It is a scenario on a model that has seen similar moves — not a 2.4 elasticity copied from a textbook.

If the model never saw 30% off

Extrapolating outside the support of history is how you discount into a hole. If you have never run a 30% fare sale in that cabin, the scenario should say so. Foundation models are not a license to invent a demand curve. They have seen other series. They have not seen your brand in a price war.

• Known-ahead flags for promos you already approved. Not same-day 'lift' leaked from actuals.
• Price as a series you will have at origin — the planned tariff, not the cleared market.
• Compete-with-yourself: a promo that steals next week is negative lag, not a win.

Lift that steals next week

Travel and grocery both hang over. Energy less so, unless you shifted load. Look at the week after, on a backtest, with the flag on. If discovery can only find “up and to the right,” ops will notice the hangover anyway — in occupancy, in pantry loading, in a dead Wednesday after a long weekend sale.

Run the path with it on and off

Change price or promo on a live Predict.ai goal and you get the new path next to the base — same model, same horizon, interval still attached. That is lift you can take to a pricing meeting. Driver discovery will have already tried the price series; if it failed the folds, the scenario should be humble.

Marketplace and internal signals (competitor fares, fuel) only help if they will exist at origin. A scraped fare you cannot rebuild at 6am is a research project, not a feature.